Nominative objects in Korean

3.1 Psychological predicates

The first type concerns psychological predicates that express a mental state. Korean psychological predicates are provided in (6), and (7) illustrates sentences with some of these predicates.^[12]

In many languages, including English (Huddleston and Pullum 2002), Russian (Croft 1993), Lakhota (Rood and Taylor 1976), and Classical Nahuatl (Andrews 1975), predicates of psychological state fall into two main classes according to the way the thematic roles align with subject (S) and object (O). In the English examples (8a) and (9a), for instance, the experiencer is aligned with the subject, while the stimulus is aligned with the object. (8b) and (9b) illustrate the opposite case.

Korean does not have experiencer-object verbs; instead, causative constructions are used to express a similar situation, as in (10).

The stimulus may also appear in the subject position through passivization, as shown in (11).

It is worth noting that the Korean NOC is neither a passive nor a causative construction. The obvious reason the NOC is not causative is the unacceptability of (12), where the putative object cannot be marked nominative. Neither does the periphrastic causative construction, shown in (13), allow a nominative-marked object.

The NOC neither permits passive morphology nor is compatible with passive predicates, as in (14a) and (14b), respectively; Mia cannot be marked nominative in the passive construction. One way to rescue (14a) and (14b) is to mark the second nominal with eykey ‘by’, as in (14c).

Another noticeable difference between the NOC and the passive construction is the nature of their subjects. While the passive construction requires a theme subject, the NOC requires an experiencer subject. To demonstrate this difference, we need to identify which nominal is the subject in the construction. The subjecthood tests for the NOC will be discussed in more detail in Section 3, but let us briefly demonstrate an example of reflexive binding. The Korean reflexive, caki ‘self’, appears to prefer the subject as its antecedent. Note that only the experiencer nominal, Gio, can be the antecedent of the reflexive caki in (15a). The second nominal, Mia, which is a stimulus, cannot serve as the antecedent of caki, as shown in (15b).

If caki is replaced with a pronoun that does not have subject orientation, it can refer to either Gio or Mia, as in (16).

The passive construction shows a different behavior. Similar to the NOC, only the first nominal, Gio, can be the antecedent of caki in (17a); here, however, Gio is a theme subject, not an experiencer or an agent. (17b) is not felicitous because caki is bound by Mia.

The examples above show that caki refers to experiencer subjects in the NOC, while it refers to theme subjects in passive sentences. In generative-linguistic analyses, a passive subject is a derived subject, and our data shows that derived subjects control the subject-oriented reflexives. In that type of approach, our observation may be interpreted as a claim that an experiencer subject with a psychological predicate is the underlying subject. While we do not subscribe to the derivational approach, and while the question of which subject is the underlying one is not relevant, the relationship between the generative linguistic notions of the derived and underlying subjects can be explained in our analysis. In accounting for the semantics of psychological verbs, Croft (1993) provides a valuable generalization, as in Figure 3. This figure illustrates that there are two processes involved in construing psychological predicates; the first involves the attention directed from the experiencer to the stimulus, and the second indicates that the stimulus causes the experiencer to enter into a certain mental state.

Figure 3:

Two-way causal relation, adapted from Croft (1993: 64).

Based on this symmetric representation, Croft makes four predictions. The most relevant one to our current discussion is summarized in (18).

(18) is predicted because the experiencer and the stimulus exhibit a two-way relation in the causal structure described in Figure 3; they are both simultaneously the initiator and the endpoint. Korean is a language that supports this prediction. The stimulus subject pattern is not available with psychological verbs in Korean. Instead, the experiencer and the stimulus are marked with the same case. Reinterpreting Croft’s generalization in CG terms, we propose that both the experiencer and the stimulus are marked nominative because each of them marks the head of a profiled event-chain.^[13]

How the linking is established between thematic roles and grammatical relations is a complex topic. Traditionally, the thematic role hierarchy is mapped isomorphically to the (underlying) grammatical relation hierarchy, and psychological predicates show some flexibility in this mapping. In this view, there is no direct correlation between case-marking and the argument structure of verbs, though some connections can be made, e.g., Burzio’s Generalization (Burzio 1987). From this perspective, Croft’s description in Figure 3 seems to be overly simplistic. However, case marking and grammatical relations do not need an intervening linking stage in many constructional approaches, including Croft’s. In CG as well, grammatical relations are determined solely based on the trajector/landmark alignment, where the subject is defined as the clausal trajector, while the objecthood is conferred on the secondary focal participant. That is, our approach does not assume any type of intermediate linking between case and grammatical relations.

3.2 The desiderative construction

The desiderative verb siph-ta also gives rise to the NOC, shown in (19); siph-ta exhibits properties of a psychological predicate because it also represents a mental state. Therefore, the situation described in (19) is stative. The difference between siph-ta and other psychological predicates stems from the former’s nature as an auxiliary verb. As an auxiliary verb, siph-ta requires a main verb.

The siph-ta construction has been discussed by many scholars, particularly within the context of case alternation; the second nominal Mia in (19) may be marked accusative, resulting in (20).

We will revisit the issues concerning the siph-ta construction in Section 6, but first, we need to address one peculiar semantic property observed in (19) and (20) here. Some extant research, such as Jung (2011), often assumes that (19) and (20) are identical in their meanings. A closer examination, however, reveals that this is not the case. As the translations of (19) and (20) indicate, (19) expresses the speaker’s mental state of missing Mia, while (20) more prominently indicates that the speaker’s desire is associated with an episodic event. The same pattern is observed with a predicate with less idiomatic connotations, as shown in (21) and (22). While (21) denotes the speaker’s mental state of craving apples, (22) is about the speaker’s mental state of desiring the episodic event of eating an apple.

The different semantic interpretations of (21) and (22) can be tested truth-conditionally. In (23), the first clause denotes the speaker’s general craving that is not anchored in a specific time. This craving can be overridden by one specific event anchored in the speech time, which is denoted by the second clause. (24) contrasts with (23) in that the first clause describes the speaker’s desire to eat apples in the speech time, which contradicts the situation described by the second clause.

Scholars such as Kim and Maling (1998) note the different semantic properties of examples like (21) and (22) and argue that the difference comes from the two distinct syntactic structures (21) and (22) manifest. Unlike Kim and Mailing, we demonstrate that the two different grammatical patterns are the realizations of two types of construals of the same conceptual content. We further show that though (21) and (22) involve the same type of complex predicate formation, they differ in their profiles. While (21) shows a construal identical to the other examples of the NOC we have discussed thus far, the profile is given to the embedded transitive relationship in (22). Case marking patterns in these examples then become merely symptomatic of two different types of construals the speaker chooses in interpreting a given situation. The details of the mechanism will be discussed in Sections 6 and 7.

3.3 A variation of the desiderative construction

The siph-ta construction discussed in Section 3.2 can be extended with the auxiliary verb ha- ‘do’, as shown in (25a). When it occurs, the second nominal cannot be marked nominative, as in (25b).

Albeit superficially similar, (25a) is different from (22) concerning agentivity and stativity. While both (21) and (22) show a lesser degree of the subject’s volitional involvement in a situation, (25a) demonstrates a higher degree of volitionality. Note that it is widely accepted that volitional involvement is a prerequisite for agenthood (see Dowty 1991; Lehmann 1991; Van Valin and LaPolla 1997; Van Valin and Wilkins 1996; Primus 1999, 2002; among others). One test for agentivity is to check the acceptability of modifying an event with adverbs indicating volitionality (see Klein and Kutscher 2002; Roeper 1987; Talmy 1976; among others).

The application of the test to examples similar to (21), (22), and (25a) is shown in (26). (25a) is still acceptable with the volitional adverb uytocekulo, as shown in (26a). (26b) and (26c), which are different versions of (21) and (22) with adverbial modification and a different verb, are not felicitous. The test results show that (25a) exhibits a higher degree of agentivity than (21) and (22). Note that the meaning of po- ‘see’ shifts to become close to ‘meet’ in (26a), due to the existence of the adverb uytocekulo ‘intentionally’. The same type of meaning shift is not available for (26b) and (26c), which again shows that the siph-e ha-ta construction forces the reading of a complex event.^[14]

The second difference concerns stativity. As described by Verhoeven (2010), a prototypical state involves no energy to go on or be kept going (also see Comrie 1976; Lehmann 1991; Van Valin and LaPolla 1997, among others). One test for stativity is incompatibility with a progressive form (see Van Valin and LaPolla 1997; Vendler 1967). The test results show that the internal structure of (27a) is non-stative, while the internal structures of both (27b) and (27c) involve stativity.

The tests demonstrate that the siph-ta construction is associated with non-agentivity and stativity, while the siph-e ha-ta construction involves agentivity and non-stativity. Since the only difference between the two constructions is the existence of the auxiliary verb ha-, it is reasonable to suggest that the function of ha- is to convert non-agentive states into agentive non-states. We propose that ha- converts an imperfective process into a perfective one, as sketched in Figure 4. In this figure, t refers to the conceived time, and the thick lines indicate the profiled portions of the process. The dotted lines in the left rectangle mean the process is unbounded, while boundedness is represented with two small vertical bars in the right rectangle.

Figure 4:

The function of the ha- verb.

CG adopts a two-way distinction of the aspectual system: perfective and imperfective. Perfectives construe the profiled relationship as internally heterogeneous, whereas imperfectives construe it as homogeneous. While heterogeneous construals involve some type of change through time, homogeneous construals involve the continuation of a stable situation through time. In this view, earlier examples like (21) and (22) illustrate stable situations of indefinite duration, in the sense that the construal excludes the beginning and the end point from the general region of attention.

The major property of states is that they exist or obtain, as opposed to occurrences, which happen (Huddleston and Pullum 2002: 119). Due to this property, a relation involving only one participant is more common for stative construals, which is the case for (21). (22) is a deviation from this prototypical stative situation because the situation describes an episodic event. Nevertheless, it can still be construed statively. It is important to emphasize that the perfective/imperfective contrast is anything but a rigid specification. A similar situation is described in (28). Though the speaker did not continuously sleep for 20 min, the first conjunct I slept for 20 min does not contradict the second clause; this situation can be viewed as a stable situation, although it shows a certain degree of heterogeneity.

Example (22) shows a case of flexible categorization of perfectivity, whereas (21) and (25a) demonstrate typical situations for imperfective and perfective construals, respectively. This type of flexibility is not a unique consideration found only in CG or cognitive linguistics in general. While not endorsing CG rationales, Hopper and Thompson (1980), Dowty (1991), and Jacobsen (1992) argue that transitivity is a scalar notion. Proponents of this view treat transitive clauses as being more or less transitive, depending on a variety of semantic factors.

According to Jacobsen (1992: 124), a highly transitive clause is viewed as involving some change of the participants, while in transitive stative clauses “the situation expressed is viewed as constituting a property of the participant itself.” Some statives can be expressed either as intransitive or transitive. When they are realized as transitive, they may take an accusative-marked object if the object is construed as affected and specific.^[15] The Korean siph-ta is one such stative predicate. Example (21) is used more naturally in the situation described in (29), in which the emphasis is on a general craving for apples. By contrast, (22) is much more natural in a situation described in (30), in which those specific apples may be affected because the speaker might purchase and eat them.

In this regard, the siph-e ha-ta construction contrasts with siph-ta in that siph-e ha-ta is construed as highly transitive, permitting only an accusative-marked object. The table in (31) summarizes our discussion on siph-ta and siph-e ha-ta. Note that we use stative tantamount to imperfective and non-stative (dynamic) to perfective. When a situation denoted by the siph-ta construction is construed as strongly imperfective, the object is marked nominative, which shows a lesser degree of transitivity. The same construction may be construed less imperfectively. While this construal still maintains stativity, the degree of transitivity is upgraded, thereby permitting an accusative-marked object. The siph-e ha-ta construction is always construed perfectively, making the choice of a nominative-marked object unviable.

The table in (31) also shows that the situation denoted by the siph-ta construction is stative without respect to its transitivity. By contrast, the siph-e ha-ta construction is strongly associated with a perfective construal, made available by the auxiliary verb ha-ta.

4.1 Subjecthood tests

There is no denying that a typical subject in Korean is marked nominative. However, the case marker itself does not determine subjecthood; there are many instances where the subject appears with non-nominative markers. One example is the dative subject, as shown in (33).

The subject of (32) is the dative-marked nominal, apeci ‘father’, as opposed to the nominative-marked nominal, ton ‘money’. The subjecthood of apeci becomes apparent because of the honorific marker -(u)si-.^[16] The inanimate entity ton ‘money’ cannot be honorified, and the only candidate for the honorification is apeci ‘father’ in (32), though it is marked dative.

We can apply the same test to the NOC to ascertain which nominative-marked nominal is a subject. When the first nominal is an entity to be honorified, the honorific marker, -(u)si-, may be affixed felicitously, as in (33). When first and the second nominals switch positions as in (34), the outcome is unacceptable. This test strongly suggests that the first nominal, not the second, is the subject of the NOC.

That said, this test needs to be used with great caution. Honorification is a discourse phenomenon, and the agreement pattern often defies a grammatical relation. For example, if the honorific marker -(u)si- agrees with a subject, the inanimate entity, ton ‘money’, must be honorified in (35); this is not a tenable explanation. Instead, we might argue that (35) is felicitous owing to -(u)si-’s function as an addressee honorifier, which is the view supported by Lim (2000). According to him, -(u)si- agrees with an addressee that can be identified in a discourse context; -u(si)- agrees with the person who drives the BMW in (35).^[17]

Without respect to the true function of -(u)si-, the test results provided thus far point to the fact that the nominative-marked NP does not exhibit any agreement relation with the honorific marker. We therefore can conclude that the second nominal in the NOC is devoid of this subject property.

The second well-attested diagnostic for subjecthood in Korean is caki ‘self’ binding, which we briefly discussed in Section 3.1: caki strongly prefers a grammatical subject as its antecedent. (36) shows that it cannot have the second nominal as its antecedent, which again demonstrates that the first nominal, not the second, exhibits a subject property.

The reflexive binding test also needs to be carefully applied. In a certain situation, caki may refer to the second nominal, as shown in (37a). If caki and the first nominal refer to the same entity, na-nun, the result is not acceptable, as seen in (37b). This is because as the third-person reflexive, caki cannot be bound by na-nun, thereby making (37b) unacceptable.^[18]

Though (37a) shows that the second nominal may bind the reflexive, its acceptability has an independent motivation. According to Kameyama (1984), a non-subject NP construed as a “perspective” can control the reflexive in Japanese. In (37a), Gio is construed as a person who suffers from the speaker hating her. Therefore, Gio is qualified as a “perspective” in this example.

The two subjecthood tests point to the suggestion that the second nominal in the NOC is not the subject, though it is realized with the nominative marker. The observation that the second nominal is not the subject does not automatically warrant that it is an object in Korean either.^[19] We already demonstrated that the nominative object exhibits different properties than those of typical objects. A typical object is marked accusative and may appear as the subject in a passive sentence. As we saw, the second nominal in the NOC is marked nominative, and the passivization of the NOC is not readily available, as illustrated in (38a)–(38c).

The discussion provided so far shows that the second nominal in the NOC exhibits in-between properties of typical subjects and objects. Incorporating intermediate properties and various degrees of prototypicality is an essential task in CG. Langacker (2008: 9) states that identifying “the range of structures that are prototypical in language as well as their degree of prototypicality” is, in fact, a primary goal of functional theory. In later sections, we account for why this type of in-between property is expected in the NOC. In sum, the second nominals in the NOC are not the subjects.

4.2 Objecthood tests

In Section 3.1, we briefly discussed that the second nominals in the NOC are not passivized, which is one potential piece of evidence for their objecthood. However, they do not exhibit a full range of object properties. Let us consider the typical tests for objecthood as shown in (39). The second nominal, Mia, cannot undergo relativization and scrambling – which are typical properties of objects – while it passes clefting, pronominalization, and wh-question tests. The outcomes of the tests illustrate that the second nominal in the NOC exhibits only partial object properties. It is important to note that these tests should not be understood as defining characteristics of objects. A nominal’s objecthood is sufficient but not necessary for these properties. In other words, objects tend to exhibit these properties, whereas exhibiting these properties does not warrant a nominal’s objecthood. The purpose of these tests here is to demonstrate that nominative objects do not even pass all of the tests often used for objecthood in the generative linguistics literature.^[20]

The tests in (39) provide additional evidence for the partial objecthood of the second nominal in the NOC; the second nominals in the NOC do not show a full range of the properties shared by typical objects. In our view, the second nominal in the NOC exhibits some object properties because it is a stimulus. At the same time, it corresponds to the internal representation of the stimulus in the domain of the experiencer’s mental experience; since the second nominal represents the head of the action-chain, it contrasts with typical objects.

4.3 The semi-copula toy-ta ‘become’

The verb toy-ta ‘become’ demonstrates some similarities to the NOC, and we need to examine if it also sanctions the NOC. Scholars such as Yoon (2007) and Levin (2017) assume that the toy- construction gives rise to the NOC because it shows stativity, and all the objecthood tests we used in (39) can be felicitously applied to that construction. Be that as it may, toy- does not show transitivity; it does not form either a transitive or an intransitive clause; rather, toy- denotes a relation of identity like the copula -i. If so, toy- behaves like a copula verb, though there are some differences between the two. While toy- in (40c) cannot be omitted, as shown in (40d), ellipsis of the true copula commonly occurs, as demonstrated in (40a)–(40b). For this reason, the verb toy- may be categorized as a semi-copula in terms of Hengveld (1992) and Butler (2003).

There is another reason we categorize toy- as a copula verb, whether it be a semi-copula or true copula verb. From a typological perspective, Dixon (2002) distinguishes copula clauses from transitive and intransitive clauses. Copula clauses have two core arguments – copula subject and copula complement – together with a copula verb. Dixon (2002: 1) provides criteria for the categorization of copula verbs. According to Dixon, copula verbs “show a relation of identity/equation or of attribution. It may also have some or all of the senses: location, possession, wanting or benefaction, and existence.” The toy- verb fits the description of Dixon’s categorization of (semi)-copula in that it shows a relation of identity or attribution, and it has two core arguments. For example, (40c) shows an identity relation between Gio and kyoswu ‘professor’, and the toy-ta verb maintains two core arguments – Gio and kyoswu ‘professor’ – without exhibiting any degree of transitivity. For this reason, we exclude the toy-ta construction from the NOC unlike scholars like Yoon (2007).^[21]

6.1 The VP-complement approach to the desiderative construction

Kim (2016: 86–89) discusses three possible analyses of the desiderative construction. The first is a VP approach, where case alternation does not entail differences of structure. This view is supported by Choi (2009) and Jung (2011). A simplified analysis of this approach is provided in (46) with the bracket notation. In these examples, siph-ta takes the embedded VP as a complement.

Kim (2016: 87–88) provides criticism of this type of analysis by pointing out that Choi’s (2009) VP analysis faces a theory-internal challenge. For example, the main verb po- ‘see’ is transitive and requires an accusative-marked object. When it occurs with siph-ta, the object may be marked nominative. This case marking pattern is a strong indication that siph-ta is responsible for the nominative marking of the object. If so, the nominative case must be assigned non-locally; siph-ta and the object belong to two different case domains. Depending on the researcher’s viewpoint, this might not be a serious issue. With some theoretical modifications, we can certainly fix the non-local case assignment problem. In fact, the same problem does not arise in a later VP complement approach, such as Jung’s (2011).

Another problem with this approach comes from examples like (47). While (47a) is acceptable as expected, the unacceptability of (47b) is not easily explained; there is no mechanism to block the occurrence of the adverb between the main verb and siph-ta in Choi’s (2009) analysis.

An even more vexing problem concerns the interpretation of (47a). When the object is marked accusative in examples like (47), the sentence denotes complex events. Since we assume that forms reflect meanings, the question of how these meanings are reflected in different case forms becomes a crucial and relevant question. In Choi’s analysis, the different semantic properties of the case marking patterns have not been addressed nor discussed. We will address this in later subsections and Section 7.

6.2 The complex predicate approach to siph-ta

The second view is a complex predicate approach, which is the view Kim (2016) adopts for his analysis. While we support Kim’s approach in the sense that the main verb and -ko siph-ta form a sort of complex predicate, we suggest that Kim’s proposal be modified, especially to account for the complex event reading we have addressed in the previous subsection.

In Kim’s approach, the main verb and siph-ta form a complex predicate in the syntax. Kim’s analysis is based on two major pieces of evidence: Negative Polarity Item (NPI) licensing and syntactic cohesion. Syntactic cohesion was already addressed in the previous section: adverbs cannot intervene between the main verb and siph-ta. Kim’s NPI test demonstrates the quintessential properties of complex predicates. Korean NPIs need to be licensed by a clause-mate negative expression. (48a) is not felicitous because the negative auxiliary verb ahn- ‘NEG’ is in a different VP domain from the NPI amwukes-to. By contrast, the negative particle, an, appears in the same clause as the NPI in (48b), yielding an acceptable result.

The same test can be extended to the siph-ta construction to check if V + siph-ta forms a complex predicate. Our prediction is borne out as shown in (49); the acceptability of (49) shows that it is indeed mono-clausal, and V + siph-ta is a single predicate.

This is a good place to identify what complex predicates are in Korean. According to Amberber et al. (2010), there is no widely accepted answer to this question nor an agreed-upon set of criteria to classify them. The researchers put forward a criterion to distinguish two types of complex predicate constructions: coverb and serial verb constructions. While coverb constructions typically involve a light verb, serial verb constructions are a sequence of verbs without any overt marker of syntactic dependency. The siph-ta construction does not fit either category; siph-ta is not a light verb, and the overt connector, -ko, is required between the main verb and siph-ta. Nonetheless, it shares fundamental properties with the serial verb construction described by Comrie (1995), which are summarized in (50).

We have demonstrated that (50a) holds true for the siph-ta construction. We have also indicated that the desiderative NOC denotes a single event, such as (I feel like) Gio misses Mia or I crave apples. The NOC examples we have discussed precisely exhibit the properties shown in (49).

The challenge is when the object of the siph-ta construction is realized with the accusative marker. Though the accusative siph-ta construction shares some grammatical properties with the nominative siph-ta construction, it appears to denote complex events as indicated with our translations like I want to eat apples and (I feel like) Gio wants to see Mia. These are perplexing examples, but we believe the complex event interpretation does not undermine the complex predicate-based analysis. Whether simple or complex events are denoted, siph-ta can form a complex predicate with the main verb. The complex predicate can then be construed either as a psychological verb or as a transitive verb that denotes complex events. In other words, one typical role of a complex predicate is to express a simple event, but the speaker can construe it as a situation denoted by a complex event. As unorthodox as it may sound, this type of availability of alternative construals might be the reason there is no consensus on the definition of the serial verb construction or complex predicates in general. It seems that there is a consensus on the mono-clausality of serial verb constructions among scholars (Butt 2010). However, the definition described in (50b) is far from settled. While Aikhenvald (2006) agrees with Comrie (1995) on the single event nature of the serial verb construction, Amberber et al. (2010) and Foley (2010) claim that the serial verb construction primarily denotes multi-events. How to define serial verbs or complex predicates in Korean is not the primary concern of our research here, but it is worth pointing out that whether or not a clause denotes a single event is not a robust measure to identify a complex predicate.

Though the complex predicate approach overcomes the challenge demonstrated in the examples in (45), it makes some similar predictions to those of the VP analysis. Since the complex predicate analysis does not entail different structures for the siph-ta construction, the nominative and the accusative cases are licensed in the same position in a simpler syntactic structure (see Kim 2016: 76). Then, under this analysis, the scopal difference between (43) and (44) cannot be explained. This is because the case alternation is accounted for by one structure, and the scopal relations among NEG, siph-ta, and ONLY are identical for both case patterns. We can avoid this problem by stating that scopal relations are completely semantic notions and are orthogonal to structure. But this is not what Kim’s analysis assumes nor pursues; therefore, Kim’s proposal needs some revisions to fully incorporate our observations.

6.3 The mixed approach

The third approach, proposed by Kim and Maling (1998), is a case of a mixed analysis. Kim and Maling posit two different structures for the case patterns of the siph-ta construction. When the object of the verb bears the accusative marking, the construction exhibits a full range of VP-complementation properties. By contrast, the siph-ta construction with a nominative-marked object does not show any of the properties of VP-complementation. For instance, Kim and Maling argue that (51a) and (51c) have a VP complement structure where two VPs are conjoined, and -ko siph is gapped in the first conjunct (see also Sells and Cho 1991). The operations are not applicable to the nominative-marked siph-ta construction as shown in (51b) and (51d).^[25] This indicates that X-ko siph-ta behaves like one unit: a complex predicate.

Though their analysis has an advantage in explaining the examples provided in (51a)–(51d), it makes some problematic predictions. Since there are two siph-ta verbs in (51a), for instance, we predict that there are two independent events. However, it is highly unlikely that we interpret (51a) as something like (I feel like) Cheli wanted to cook rice and wanted to do the laundry (not necessarily in that order). To us, only the sequential reading is available: (I feel like) Cheli wanted to cook rice and do the laundry (in that order). If that is the case, what siph-ta takes as the first conjunct is not a full VP, and the conjoined structure has only one ‘desiring state.’

A more serious issue with Kim and Maling’s analysis is the increased acceptability of examples like (51b) with an appropriate choice of verbs coupled with relevant pauses, as shown in (52). Here, both // and /// represent pauses, where /// notates a longer pause than //.

Kim and Maling’s analysis predicts that (52) must be unacceptable; since mek-ko cannot form a complex predicate with siph-ta in the first conjunct, the nominative marking is not allowed. Contrary to this prediction, (52) is natural in casual conversation, which weakens the validity of their dichotomous structure-based claim.

Despite some weaknesses, Kim and Maling provide an insightful analysis of the siph-ta construction. In particular, their argument that the single-event and the multi-event readings are associated with different structures resembles the proposal we put forward, though our focus is on semantic structures. That being said, our analysis takes the opposite direction; while Kim and Maling assume that the structural difference leads to different interpretations of the siph-ta construction, we show that the different case patterns are a mere reflection of the two types of conceptualization that the siph-ta construction motivates. A detailed discussion of our analysis is provided in Section 7.

6.4 On case alternation and different types of construals

We have thus far made two assumptions in dealing with meanings. First, the semantic structure is determined based on the interlocuter’s construals. Second, grammatical forms determine meanings. As for the first assumption, Croft (2012) characterizes a construal as a conceptual semantic structure, as in (53).

The characteristics provided in (53) accurately summarize our view. Alternative construals are ubiquitously observed in everyday language use. The English verb see is often categorized as a state verb in Vendler’s classification (Vendler 1967) because the situation described by the verb does not change over time and does not have a natural endpoint. However, the same verb may have alternative construals as in (54a) and (54b). Without any morphological derivation, the situation described by the verb can be construed as a transitory state as in (54a), or the verb may have an achievement construal as in (54b) with the past morphology.^[26]

Similar types of alternative construals are available in Korean. The verb a(l)- ‘know’ that is typically analyzed as a state verb may have alternative construals. While (55a) shows an example of a transitory state construal, (55b) shows an achievement construal. Note that the a(l)- verb in (55b) is interpreted as past, though there is no past marking.

Working from this perspective, the suggestion that case alternation is an outcome of alternative construals is expected. In dealing with adverbial case marking in Korean like (56), Park (2013) argues that accusative marking is a result of the predicate’s perfective construal of a situation, while the nominative-marked variety arises when the same predicate is construed imperfectively. For example, (56a) is understood as a continuation of an ongoing stable situation. In (56b), however, the raining event is internally heterogenous, resulting in the interpretation of undirected activity in terms of Talmy (1985).^[27]

More specifically, Park argues that the case alternation is tied to how the speaker construes a given situation. When the situation is construed imperfectivly, the adverbial functions as a global setting that includes a whole event. In this situation, it is marked nominative. By contrast, when the situation is construed perfectively, the adverbial characterizes a fragment of a setting, i.e., a location, which is the site of a single participant. In this instance, it is marked accusative.^[28] We will not go into detail on Park (2013), but his analysis demonstrates how different construals lead to different case markings, thereby showing how our two assumptions – construal and form/meaning connection – are satisfied in analyzing case alternation phenomena.

From non-cognitive viewpoints, Kim and Sells’s (2010) and Lee’s (2017) analyses of Korean adverbial case also strongly demonstrate that a syntactic account alone is not satisfactory to understand case alternation phenomena. These researchers emphasize the importance of the predicate’s aspectual type and the degree of animacy of the subject. They go even further to show that aspectual types and the degree of animacy might vary depending on the given context, or depending on the speaker’s construal if we rephrase their arguments.^[29]

Turning back to the case alternation in the siph-ta construction, we propose that the complex predicate, V + siph-ta, may be construed as a psychological predicate, which yields a nominative-marked object. Alternatively, the same complex predicate may be construed as a complex event. One curious example is shown in (57). At first glance, it seems that po-ko siph-ta cannot have a psychological verb construal due to the existence of the manner adverbial ppali ‘quickly’. Even then, the object is marked nominative.

Note, however, that the adverb ppali ‘quickly’ is polysemous, and it has the meaning of an intensifier, as indicated by the translation in (57). So (57) is identical to (58) in its intended meaning, and the nominative marking on the object is expected.

Thus far, we have identified major properties of the Korean NOC. The next section provides our technical CG analysis of the NOC.

7.1 NOCs with psychological verbs

We begin with the NOC with simple psychological predicates. The CG diagrams for (60) are provided in Figure 5.

Figure 5:

Third-person versus first-person NOC.

Figure 5(a) is for (60a), where the speaker (S) takes the viewpoint of the third-person experiencer (C′) by creating a surrogate scope (SS). The stimulus, Mia, sends a stimulative signal to the surrogate experiencer, Gio.^[31] Gio then undergoes an internal change of state, and Mia existing in the CS is a part of the state resulting from the change. Gio experiences this state of mind by making mental contact with it, indicated by the dashed arrow within the largest circle. Note that (61a) is construed as the subject’s experience assessed by the speaker. (60b) is essentially identical to (60a) except for one component. In (60b), the conceptualizer and the surrogate conceptualizer are identical; therefore, the two scopes collapse, yielding the semantic structure in Figure 5(b).

As for the case marking pattern in (60a), Mia is the sole participant in the relationship described within the surrogate scope and is the head of the action-chain; therefore, it is marked nominative. The surrogate experiencer, Gio, is the primary participant in the profiled process, which makes mental contact with the state of mind. Therefore, it is marked nominative as well.

7.2 The NOC with the verb siph-ta

The major issue concerning the desiderative construction revolves around the case alternation the construction allows. We reintroduce the alternation in (61).

(61a) and (61b) are illustrated in Figure 6(a) and (b), respectively. Let us first discuss Figure 6(a) where the thick lines indicate the profile. The internally represented stimulus, Mia, is not a participant in the profiled process but a part of the embedded relationship, which is enclosed within the dashed rectangle. The verb siph-ta profiles both the existential relationship denoted by the solid-line arrow and the experiential relationship denoted by the dashed-line arrow. The former represents the existence of the transitive process within an emotional domain in the mind of the experiencer. The zigzag line inside the dashed rectangle represents the transitive relationship, po-ko ‘see-COMP’. In the embedded relationship, the trajector of the transitive relationship is only schematically represented and is identified as the surrogate experiencer (C′) by the correspondence relation indicated by the dotted line. Within the surrogate scope, Mia is the sole participant, thereby gaining the trajector status and being marked nominative. The semantic structure for (61b) is essentially identical to that of (61a), with the only difference manifesting in the profile; the profile is given to the embedded transitive relationship in (61b), and all the outer layers pertaining to experience are in the background and out of profile, as indicated by non-bold lines.

Figure 6:

Two types of [V + siph-ta] constructions, based on Kumashiro (2016: 219).

Our analysis makes several desirable predictions concerning the desiderative construction. The first concerns the case-marking patterns. When siph-ta is construed as having the semantic structure in Figure 6(a), the object nominal is realized with the nominative marker because the new trajector status is conferred on the nominal. Figure 6(b) shows how the accusative marking arises. In the embedded transitive relationship, the active participant of the profiled process is Gio, and the passive participant is Mia. Therefore, the nominative and accusative markings are given to them, respectively.

One might view the structure in Figure 6(b) similar to the VP complement analysis proposed by Kim and Maling (1998). Our analysis is similar in that the embedded clause is required by siph-ta, which leads to an interpretation similar to the VP complement structure found in Kim and Maling. The noticeable difference is we do not have to compose the full VP first in the structure shown in Figure 6(b). For example, X-ko siph-ta is composed with the verb po-ta ‘see-DECL’ to create a complex predicate, followed by the elaboration of the object and the subject.^[32] Our analysis systematically accounts for coordination and gapping upon which Kim and Maling rely to support their analysis. (62a) is felicitous, where two siph-ta constructions are conjoined. When the first conjunct is used without -ko siph as in (62b), the most natural interpretation is to view the event as a sequenced one. The interpretation indicated in (62c) is only marginally acceptable, or not available at all.

Kim and Maling’s analysis cannot explain the (62c) interpretation easily because the first VP conjunct is fully loaded with -ko ‘COMP’ and -siph, followed by the gapping of these elements. Then, we expect to have the intended meaning without difficulty because there are two independent ‘desiring states’ denoted by two instances of siph-ta. The same issue does not arise in our analysis because there is only one ‘desiring state’ in the siph-ta construction. Therefore, examples like (62c) are interpreted as one ‘desiring state’ of a sequence of events, as opposed to desiring two independent events.

The other noticeable difference between our analysis and Kim and Maling’s concerns the composition of the siph-ta construction. Kim and Maling propose that there are two distinct compositional structures that are held accountable for two case markings. As discussed previously, we are not positing two discrete structures for two different case markings. Both of the case marking patterns come from the same semantic base; the difference is which portion is profiled. Our analysis also correctly predicts the unacceptable coordination of the NOC as shown in (63b). (63a) is acceptable simply because we conjoin the two identical structures shown in Figure 6(a). Our analysis does not permit (63b) because the main verb po- ‘see’ requires siph-ta to allow a nominative marking on the object. There is only one siph-ta in (63b), which forms a complex predicate with mek-ko ‘eat-COMP’, leaving the nominative-marked object in the first clause unaccounted for. As we discussed in Section 6.3, (63c) is acceptable, although it exhibits a similar syntactic structure to (63b). This is because pap-i-mek-ko and maycwu-ka masi- conjoin first followed by the -ko- siph-ta attachment. Though the same type of coordination pattern is certainly possible for (63b), it is undesirable because it is difficult to see the conceptual connection between missing Mia and craving apples to create one juxtaposed unit. In CG, coordination is characterized as the mental juxtaposition of entities conceived as being analogous (Langacker 2012). From this perspective, the (un)acceptability of (63b) and (63c) is fully expected due to the varying degrees of mental juxtapositionality of their components.

The third prediction our analysis makes concerns the episodic reading the structure in Figure 6(b) permits. The same reading is not readily available for the structure in Figure 6(a). (64a) represented by Figure 6(a) is interpreted as ‘I missed Sam continuously during the whole year.’ The same reading is much weaker or not readily available in (64b), which is represented by Figure 6(b). Instead, the prominent reading of (64b) is an episodic reading, such as ‘I wanted to see or meet Sam sometime during my stay in Quahog (maybe by chance).’

The meaning difference in (64a) and (64b) is prominent due to the clear polysemous property of the verb po-ko siph-ta; it can either mean ‘miss’ or ‘meet/see’. But a similar type of meaning difference is observed with verbs with less polysemous usages as well. The NOC (65a), which has the same structure as (64a), is interpreted as describing a more continuous status of desiring for the study of linguistics, e.g., majoring in linguistics. With the second nominal marked accusative, as in (65b), the meaning becomes less permanent, e.g., studying linguistics without continuous commitment.

While the difference shown in (65a–b) is subtle, the examples illustrate that the situations described by the two are not identical.

The last prediction our analysis makes concerns the scope of the nominative-marked object. In a typical situation, it takes wide scope in relation to negation and siph-ta. Our analysis naturally predicts the scope-related interpretations. In the desiderative NOC, as depicted in Figure 6(a), the trajector status is conferred on the passive participant, which exhibits a higher degree of non-sentential topicality. Owing to this higher degree of topicality – albeit non-sentential – the nominative-marked object poses wide scope.^[33] Note that topicality should be understood on a spectrum. While a reference point trajector typically exhibits the highest degree of topicality, a non-reference point trajector like a nominative object also exhibits a certain degree of topicality in relation to landmark or negation. This is the reason why nominative objects can often be interpreted contrastively. Variations of the scopal relations arise when a given situation is construed differently. For example, an accusative-marked nominal can take wide scope in relation to negation or siph-ta as we discussed with (45a). The reason for this particular variation is the topical nature of the accusative-marked nominal. We discussed that the wide-scope interpretation of an accusative-marked nominal in the desiderative construction arises when it is interpreted as a sentential topic.

8.1 Major subject

It is worth discussing whether the nominative object is a major subject, which is supported by Saito (1982). Both Takano (2003) and Koizumi (2008) provide arguments against the major subject approach to the Japanese NOC, but their criticism is not extensive, and Korean exhibits different patterns. We provide our evaluations of Takano (2003) and Koizumi (2008) in the next subsection, but first let us focus on their argument against the major subject analysis.

In order to discuss their argument, we need to understand what major subject is. One of the most thorough studies on major subjects in Korean is found in Yoon (2007). Yoon’s interest is to account for Subject-to-Object (SOR) raising using the notion of major subject. The schematic structure of Yoon’s major subject is depicted as the tree diagram in Figure 7. In this type of analysis, there are always two subject positions in a clause: Major Subject (MJS) and Grammatical Subject (GS). WP is the landing site of the MJS in the matrix clause, and GS and the predicate form a Sentential Predicate.^[34] Here, categorical and thetic subjects refer to the topical and the non-topical subjects, respectively.

Figure 7:

A schematic structure of major subject in SOR, adapted from Yoon (2007: 633).

Yoon (2007: 637) also provides six interpretive properties of raised nominals in the Korean SOR construction. According to him, these are all symptomatic of the raised nominals’ origin; they were major subjects in their respective embedded clauses. We will not discuss the properties in detail here because most of them are not directly related to the present article, but we take into account one of Yoon’s interpretive properties in (66) to see if a nominative object is indeed a major subject.

If the first nominal in the NOC is a major subject, we expect that it exhibits the interpretive property laid out in (66). This prediction is not borne out. As illustrated earlier, the psychological predicates that sanction the NOC are often construed as stage-level predicates. The other interpretive properties of nominative objects that Yoon discusses but that are not introduced here are available only when SOR itself is permissible. Put differently, according to Yoon, major subjects can undergo SOR. Unlike typical examples with a major subject, the SOR operation is not permitted with the NOC.

Let us consider (67a), which contains an embedded NOC. If the first nominal in the NOC, na ‘I’, is a major subject, it should be able to undergo SOR, which would yield (67b).

However, (67b) is not felicitous. Our observation demonstrates that the nominative object does not share the properties major subjects exhibit.

Another problem with the major subject approach concerns the properties of a sentential topic. Though Yoon (2007) does not directly address this issue, his arguments strongly allude to the idea that the major subject is a sentential topic. The sentential topic nature of the major subject is more thoroughly examined in Kumashiro and Langacker (2003) for Japanese and in Park (2019) for Korean. These scholars use the term reference point subject to refer to a major subject. Since reference point “invokes the conception of one entity in order to establish ‘mental contact’ with another” (Langacker 2008: 83), the sentence-level topicality of a reference point subject is expected. In the double nominative construction (68), the major subject, Gio, as a reference point subject, naturally denotes sentence-level topicality.

The same type of topicality is maintained in an embedded clause as in (69a) and after the SOR operation shown in (69b).

Unlike major subjects, the first nominals in the NOC do not denote sentential topicality.^[35] A simple NOC like (70a) may be interpreted as a plain or a contrastive statement. The siph-ta desiderative construction exhibits the same property as in (70b).

To acquire the topic status, the first nominals in (70a–b) need to be explicitly marked with the topic-marker as in (71a) and (71b).

The data and tests presented above show that the major subject approach is not a viable option for analysis of the Korean NOC. Therefore, we need a separate treatment of the NOC from the major subject construction.

8.2 Proleptic object

The second approach to the Japanese NOC is Takano’s (2003) proleptic object analysis. His analysis treats the nominative object similar to the English proleptic object as in (72). The proleptic object, John, in (72) is assumed to be base-generated outside the embedded clause and binds the pronoun, he, in the embedded clause.

Takano (2003) suggests that the nominative object is the object of the matrix clause; it is neither a major subject nor a thematic subject. So, in (73), the nominative object, doitugo-ga ‘German-NOM’, is base-generated as an object of the matrix verb in the specifier position of the higher verb.

The tree structure for (73) is provided in Figure 8. Note that the structure is bi-clausal, and the nominative object binds the pro object in the embedded predicate (V2).^[36]

Figure 8:

A proleptic object of the Japanese NOC, from Takano (2003: 795).

Takano provides a robust analysis of the given data within the generative linguistics framework. The purview of his research is to analyze the Japanese NOC with complex predicates by positing a bi-clausal structure. Takano puts forward a reasonable proposal for NOCs with complex predicates, though he does not discuss the NOC with simple psychological verbs. Since they are mono-clausal, we cannot postulate a proleptic object, which is an object of the embedded verb. For the same reason, we also cannot identify the binding relation between the nominative object and pro in the mono-clausal NOC. It is certainly possible to account for the rise of nominative-marked objects in a mono-clausal structure with some modifications of Figure 8. For example, according to James Yoon (personal communication), one way to resolve that issue is to assume a decompositional syntax for a simple psychological predicate, e.g., V-v. In this structure, the nominative case may be assigned to the object due to its feature [+ stative]. While this type of analysis might be acceptable within the framework Takano adopts, we have no choice but to ask if this analysis can really explain why objects are marked nominative with certain types of stative verbs.

Takano’s analysis faces a theory-internal challenge even dealing with examples with complex predicates. Koizumi (2008: 150) provides a set of examples to show a scopal contrast. While migime-dake ‘only his right eye’ has a narrow scope in (74a), (74b) illustrates the case where it renders a wide scope reading.

According to Koizumi, Takano’s analysis fails to predict the interpretation shown in (74b). If a nominative object is licensed by a stative verb described above, it needs to be licensed within the embedded VP as shown in (75). Since the nominative object is asymmetrically c-commanded by the higher verb, we expect that the nominative object cannot take scope over Neg. Contrary to the prediction, the nominative object has a wider scope in comparison to Neg as in (74b), making Takano’s proposal incomplete.

Koizumi’s analysis (Koizumi 1994, 1995, 2008) overcomes the issue with Takano’s proposal. In Koizumi’s proposal, the nominative object undergoes movement to the domain of Tense (T) for the purpose of case licensing – the nominative case is uniformly licensed by Tense regardless of which NP it appears on. Since the domain of T is higher than Neg, the nominative object in (75b) has a wide scope reading.

Be that as it may, just like the case of Takano’s analysis, Koizumi’s analysis does not rely on the interaction between the experiencer and the stimulus, which we believe is a pivotal property of the NOC. The scopal difference between (75a) and (75b), which Koizumi uses to demonstrate the superiority of his analysis to Takano’s, is not foolproof either. As discussed in Section 5, whether a nominal takes a wide or narrow scope is often more flexible than many researchers have observed.