2.1. Nominative object constructions in Korean

Nominative object constructions (NOCs) represent a distinctive type of double nominative constructionFootnote ² found in some languages, including Korean. In Korean, the subjects of both transitive and intransitive verbs are typically marked by the nominative case marker -i/ka, while objects of transitive verbs are marked by the accusative case marker -(l)ul, as shown in (7). While Korean sentences often have a nom-acc case frame, certain predicates – particularly those associated with experiencer psych verbs – allow two nominative-marked NPs within the same clause, creating structures with two clause-mate nominative NPs, as in (8).

This configuration features an experiencer subject and a theme object, both marked with nominative case. Notably, nominative object constructions are restricted to predicates with experiencer subjects; we do not find NOCs with other types of subject roles. Nominative objects behave as syntactic objects rather than subjects, as they do not trigger honorific agreement on the verb. The honorific verbal affix -si-, which reflects the speaker’s respect toward the subject, is only triggered by the first nominative nominal in NOCs, as shown in (8b). This indicates that this NP is formally the subject. However, in (8c), where honorific Teacher Kim is the nom-marked object, the verb cannot bear the honorific marker. In (8d), the nominative case marker appears on the subject NP to the right of the dative marker -eykey, a phenomenon called ‘case stacking’. Case stacking in Korean, where a nominal bears multiple markers, has been analyzed as combining abstract case with discourse-related markers such as topic or focus (see, e.g., Gerdts & Youn Reference Gerdts, Youn, MacLeod, Larson and Brentari1988, Yoon Reference Yoon, Bhaskararao and Subbarao2004, Lee & Nie Reference Lee and Nie2022). A detailed analysis of case stacking is beyond the scope of this paper, but we consider case stacking once again in our Conclusion section.

Constructions like these appear to violate the Distinctness Principle (Bornkessel-Schlesewsky & Schlesewsky Reference Bornkessel-Schlesewsky, Schlesewsky and Van Valin2008, Bornkessel‐Schlesewsky & Schlesewsky Reference Bornkessel-Schlesewsky and Schlesewsky2009). In terms of mental representation, this principle posits that referents in any given event should be as different as possible from each other. In terms of the linguistic form of a sentence, this means that any two clause-mate NPs should be as morphologically and functionally distinct as possible. These observations raise important questions about what factors license or constrain nominative object constructions in Korean. To address these issues, we next introduce the theoretical background on accessibility hierarchies, which will provide a framework for analyzing the well-formedness of these constructions.

2.2. Accessibility: Grammatical function, thematic role, and morphological case

In a broad sense, ‘accessibility’ refers to how salient or prominent an NP is in any given sentence. Arnold (Reference Arnold2010: 188) defines accessibility as ‘the property of information that makes it easier to access, independently of ambiguity considerations’; that is, an NP is deemed as more accessible than other NPs in the same sentence if it is the most salient NP in mental representation. Based upon typological observations about syntactic and morphological patterns across languages, three core hierarchies have been proposed (e.g., Keenan & Comrie Reference Keenan and Comrie1977, Comrie Reference Comrie1989, Aissen Reference Aissen2003), in which NPs are ranked according to how accessible they are.

The first of these is the Accessibility Hierarchy of grammatical function (Keenan & Comrie Reference Keenan and Comrie1977). Based on a survey of relative clauses in a subset of the world’s languages, these authors propose that subject is the most accessible grammatical role, followed by object, and then by indirect objects and more peripheral NPs. This hierarchy is shown in (9).

According to Keenan and Comrie (Reference Keenan and Comrie1977), the ranking in (9) is operational across languages in the sense that subject relative clauses (e.g., The reporter [who __ _subject saw the senator].) are the most common kind of relative clauses cross-linguistically. If a language allows object relative clauses, then it necessarily also allows for subject relative clauses, but not vice versa. The hierarchy in (9) is also adopted in psycholinguistic literature to account for behavioral observations about how language users produce and interpret referring expressions such as pronouns (Gordon et al. Reference Gordon, Grosz and Gilliom1993, Grosz et al. Reference Grosz, Joshi and Weinstein1995).

The second type of hierarchy concerns the thematic role of an NP (e.g., as an agent, experiencer, theme, goal, etc.). The thematic role hierarchy ranks thematic roles based on their prototypical properties, with proto-agents (such as agents and experiencers) being more prominent than proto-patients (such as themes and goals) (e.g., Fillmore Reference Fillmore1968, Jackendoff Reference Jackendoff1972, Givón Reference Givón1984, Dowty Reference Dowty1991, Primus Reference Primus1999, Grimshaw Reference Grimshaw1990, Speas Reference Speas1990, Van Valin Reference Van Valin and Robert1990), as shown in (10). Across languages, a thematic role at the left edge of the hierarchy (e.g., agent) gets mapped to subject position. A role toward the right edge (e.g., patient) usually gets mapped to the object role (Dowty Reference Dowty1991).

Related to this, certain semantic features such as animacy and definiteness can make NP a better or worse fit for one specific type of thematic role compared to other roles; for example, an NP that is animate and human is more likely to get interpreted by an addressee as being a proto-agent (e.g., Rissman et al. Reference Rissman, Woodward and Goldin-Meadow2019) because such properties make it a good fit for this role (Silverstein, Reference Silverstein and Dixon1976). In this same vein, an inanimate NP is more likely to be interpreted as a proto-patient. In fact, NPs whose semantic properties are misaligned with their expected proto-role tend to be overtly case-marked in languages, a phenomenon known as differential argument marking (e.g., Moravcsik Reference Moravcsik, Greenberg, Ferguson and Moravcsik1978, Comrie Reference Comrie1979, Bossong Reference Bossong1985, Comrie Reference Comrie1989, Aissen Reference Aissen2003).

The third hierarchy we discuss is the morphological case hierarchy (Otsuka Reference Otsuka, Johns, Massam and Ndayiragije2006, Bobaljik Reference Bobaljik, Harbour, Adger and Béjar2008). This hierarchy concerns the morphological form that an NPs surfaces with, and it ranks NPs based on their case features. NPs that have so-called ‘unmarked’ case – nominative or absolutive – are more accessible for formal syntactic operations like A- and Ā-movement compared with NPs with marked case (e.g., ergative, accusative, dative), which are less accessible. For example, in some languages, only nominative or absolutive NPs can be Ā-extracted, and other NPs that have ergative, accusative, or dative case cannot be (e.g., Aldridge Reference Aldridge2004). These observations motivated Otsuka’s (Reference Otsuka, Johns, Massam and Ndayiragije2006) morphological case hierarchy, shown in (11).

Putting these three hierarchies alongside each other leads to the hypothesis that the most accessible types of NPs across languages are unmarked proto-agent subjects, whereas the least accessible are oblique-marked, proto-patient objects.

What is important is that grammatical role, thematic role, and morphological case do not necessarily align, meaning that subject NPs are not always unmarked (as in ergative-absolutive languages, for example) and/or not necessarily proto-agents (as in passive and unaccusative constructions). Likewise, object NPs are not always marked (as in Korean sentences like in (4) and (5)), and they are not always proto-patients (as object experiencer constructions; Landau Reference Landau2010). What happens in such circumstances? To date, much literature has focused on instances of misalignments between grammatical function and thematic role, as can be the case with so-called psych predicates (e.g., Belletti & Rizzi Reference Belletti and Rizzi1988, Bornkessel & Schlesewsky Reference Bornkessel and Schlesewsky2006, Landau Reference Landau2010, Wagers et al. Reference Wagers, Borja and Chung2018, Wilson & Dillon Reference Wilson and Dillon2022). Psych predicates include English verbs like delight, admire, dislike, and irritate, where one of the two NP arguments is a thematic experiencer, in the sense the referent is the bearer of a mental state. For psych verbs like admire and dislike, the experiencer is mapped to subject position (called subject experiencer verbs). But for others, like delight and irritate, it is the object that is the experiencer (called object experiencer verbs). As recently discussed by Wilson and Dillon (Reference Wilson and Dillon2022), object experiencer verbs trigger a sentence where the grammatical function hierarchy (Subject > Object) and thematic role hierarchy (Experiencer > Theme) are misaligned, because the subject is the theme, and the object is the experiencer. By contrast, subject experiencer verbs exhibit a prototypical alignment between grammatical function and thematic role (subject = experiencer; object = theme). Indeed, Wilson and Dillon (Reference Wilson and Dillon2022) examined how sentences with object experiencer verbs are processed compared with subject experiencer counterparts. In two experiments, they found that misaligned object experiencer configurations are more costly to process compared with aligned subject experiencer ones. This indicates that misalignment of these two accessibility hierarchies is associated with comprehension difficulty on the part of a language user.

With these observations in mind, we now discuss two types of constraints which are relevant for understanding the source of misalignment. The first is local constraints, which pertain to properties of individual arguments. The second is global constraints, which concern the overall configuration of arguments within the clause, and which penalize certain combinations of subject and object case marking.

2.3. Local constraints: Rarity of nominative objects

Nominative objects are relatively rare, cross-linguistically (Silverstein Reference Silverstein and Dixon1976, Keine Reference Keine2010, Malchukov Reference Malchukov2010, Georgi Reference Georgi, Alexiadou, Kiss and Müller2012, Clem & Deal Reference Clem and Deal2024): Nominative case is canonically associated with subjects or agents. Influentially, Silverstein (Reference Silverstein and Dixon1976) introduced the referential hierarchy, on which nominative (or unmarked) case is typically reserved for high-ranking arguments – most prototypically agents/subjects and/or speech-act participants (i.e., 1st or 2nd person NPs). Conversely, when low-ranking arguments (e.g., inanimate or indefinite NPs) appear in subject position, or when high-ranking arguments appear as objects, languages often employ overt or differential case marking to signal this type of “atypical” configuration. Thus, assignment of nominative case to objects, especially those low on the referential hierarchy, is typologically rare (because high-ranking arguments indeed are generally mapped to subject position).

This rarity has been linked to broader cross-linguistic constraints on case alignment and argument structure. One key issue is that nominative objects can disrupt the expected associations between case and argument thematic roles, reducing the morphosyntactic distinctiveness between subject and object. From a syntactic perspective, it has been argued that the presence of nom case on non-subjects (e.g., objects or embedded subjects) is constrained by locality conditions and structural licensing mechanisms (Keine Reference Keine2010). These mismatches, particularly when nominative is assigned to a theme or patient, frequently demand repair strategies like movement or restructuring (see Keine Reference Keine2010, Malchukov Reference Malchukov2010, Georgi Reference Georgi, Alexiadou, Kiss and Müller2012), or sometimes cannot be saved at all (e.g., Clem & Deal Reference Clem and Deal2024).

2.4. Global constraints: Alignment and distinctness

In addition to local constraints on the types of arguments that get nom case, sentence acceptability is shaped by global, clause-level constraints. According to the Distinctness Principle (Bornkessel-Schlesewsky & Schlesewsky Reference Bornkessel-Schlesewsky, Schlesewsky and Van Valin2008), two clause-mate NPs should be as morphologically and functionally distinct as possible: When two arguments within a clause are marked with the same case (such as in a nom-nom configuration), this reduces their morphosyntactic distinctness and is expected lead to increased processing difficulty or reduced acceptability.

This principle is closely related to the broader concept of alignment, which concerns the mapping between grammatical functions (subject, object), thematic roles (agent, patient, experiencer), and morphological case. In canonical nominative–accusative alignment systems, subjects of both transitive and intransitive verbs are treated alike and are distinct from objects, typically through case marking or agreement patterns. When alignment between these domains is disrupted: for example, when both subject and object receive nominative case, this creates ambiguity in argument structure and undermines the functional distinctness that case marking can provide.

In our forthcoming Morphological-Thematic-Grammatical alignment hypothesis, morphological case marking plays a central role: For example, in nom-acc constructions (as in (7b)), the subject (agent or experiencer) is marked with nominative case and the object (patient) with accusative case, preserving clear distinctions across all three domains. In contrast, double nominative constructions (nom-nom) or mixed patterns (dat-nom) disrupt this alignment, as the mapping between case, grammatical function, and thematic role is not preserved. Thus, both local constraints (e.g., the atypicality/unexpectedness of a nom-marked object) and global constraints – particularly those formalized by the MTG alignment hypothesis – weigh in to influence sentence acceptability. In our forthcoming study, we show that both levels of constraint are operative. We also propose that morphological marking of both the subject and the object, as captured by the MTG, is influential in shaping the acceptability and distribution of nominative object constructions in Korean. These local and global constraints can be further illuminated by considering broader patterns of accessibility and alignment in language.

2.5. Interim summary: Distinctness, alignment, and accessibility

To briefly summarize our opening discussion so far, we have noted that nom-nom constructions, like the Korean example in (4), appear to violate two principles of language well-formedness. First, at the local level, the assignment of nominative case to an object is unexpected (Silverstein Reference Silverstein and Dixon1976, Keine Reference Keine2010, Malchukov Reference Malchukov2010, Georgi Reference Georgi, Alexiadou, Kiss and Müller2012, Clem & Deal Reference Clem and Deal2024). Second, at the global level, NOM object constructions are predicted to disrupt expected alignments between case, grammatical function, and thematic role. Looking back at the Korean sentences in (4) and (5), we note that the nom-nom configuration violates the expectation that clause-mate NPs should be morphologically distinct (since both subject and object bear nominative case) and the expectation that case marking should consistently map onto grammatical and thematic hierarchies (since the lower-ranked object/theme receives the higher-ranked, nominative case).

Our experimental investigation tests how Korean speakers respond to these configurations. The first goal of our study is to establish how speakers of Korean respond to nom-nom and dat-nom configurations, vis-à-vis configurations that satisfy both distinctness and alignment; thus, nom-acc configurations. At the same time, however, grammaticality is of course not determined by alignment or (a)typicality alone: Certain NPs simply cannot be nominative. Specifically, the Korean light verb -ha- is obligatory for licensing accusative case in nom-acc constructions (Jung Reference Jung2002, Reference Jung2003, Yeo Reference Yeo2004, Choi Reference Choi2019, Choi Reference Choi2022). Without -ha-, even an apparently aligned NOM-ACC configuration is ungrammatical (e.g., paywu-ka (nom) kamtok-ul (acc) coha-*(ha)-yess-ta). This shows that local syntactic constraints (e.g., the Korean nom-case licensing rule) can block otherwise well-aligned case arrays. Thus, while alignment and markedness shape acceptability, they operate within the bounds of other language-specific morphosyntactic requirements. We put these matters to the test with a novel acceptability judgment study.

3.1. Method

3.1.2. Materials

The experiment employed a 3 × 2 within-subjects design, manipulating two factors: case (nom-acc, nom-nom, dat-nom) and light verb (light verb, no light verb). We created sentences like those in (12), both with and without light verb forms, in nom-acc, nom-nom, and dat-nom case frames. An example item set is shown in Table 1.

Table 1. A sample set of target items

A total of 24 experimental trials were created, with each trial consisting of three words in SOV order: An animate subject, followed by an animate object, followed by an experiencer verb. Six lists were generated using a Latin square design, ensuring each participant saw each item only once. Participants saw a total of 4 items per condition over the course of the experiment, yielding a total of 192 tokens per condition (cf. other recent acceptability judgment experiments on Korean, e.g., Kim et al. Reference Kim, Kwon, Lee, Lee, Lee and Yoon2021).Footnote ⁴ Additionally, 24 fillers were included, consisting of intransitive sentences, of which half were unacceptable.

3.2. Results

The analysis included all 1152 critical trials. The raw acceptability ratings in each condition are shown in Figure 1. Qualitatively, we found that sentences with nominative objects (nom-nom and dat-nom) were rated lower in acceptability than sentences with accusative objects (nom-acc). Additionally, sentences containing a light verb were rated more acceptable than those without one in nom-acc and nom-nom, with the largest improvement observed in the nom-acc condition. The fully acceptable fillers received a mean rating of 5.49 (SD = 2.14), while the unacceptable fillers received a mean rating of 2.73 (SD = 2.09).Footnote ⁵

Figure 1. Mean ratings for each level of case – nom-acc, nom-nom, and dat-nom – for each level of light verb (light verb in plain bars; no light verb in patterned bars). Error bars indicate 95% confidence intervals.

We analyzed ratings by fitting a 2 × 3 cumulative link mixed-effects regression model with crossed random effects for participants and items (Baayen, Davidson & Bates Reference Baayen, Davidson and Bates2008; Jaeger Reference Jaeger2008), as implemented in the Ordinal package of the statistical software R 4.3.2 (R Core Team 2015; Christensen Reference Christensen2023). The predictor light verb was sum-coded (coefficients of 1/2 for light verb and –1/2 for no light verb). We coded the 3-way case predictor using centered Helmert Contrasts, as this coding scheme best reflects our theoretical questions. Our first contrast, called nominative object, compared the nom-acc conditions (Coefficient: 2/3) with the nom-nom and dat-nom conditions, pooled (Coefficient: –1/3 for each level). This contrast asks about how ratings of the sentences are affected by the presence of a nominative-marked object. Specifically, it tests for the role of cross-hierarchical alignment: Both nom-nom and nom-dat constructions are misaligned across the three accessibility hierarchies that we introduced in Section 2, whereas nom-acc is aligned. The second contrast, called double nominative, compared the nom-nom condition (Coefficient: 1/2) with the dat-nom condition (Coefficient: –1/2). This contrasts tests for role distinctness by comparing sentences with a non-distinct case array (nom-nom) with those with a distinct one (dat-nom). The coefficient for nom-acc in this contrast is 0, because it does not participate in the comparison. We report statistical outputs for the model with the full, maximal random effect structure supported by the design (Barr et al. Reference Barr, Levy, Scheepers and Tily2013). The output of the model is shown in Table 2.

Table 2. The 2 × 3 mixed-effects cumulative link regression model (top part), along with the effect of Light Verb at each level of Case (bottom part). Significant effects are bolded.

We find first, a significant main effect of light verb, with sentences with a light verb rated as higher on average than those without (means = 3.8 vs. 2.5; β = 2, SE = .23, z = 8.9, p <.0001). As indicated in the graph, this main effect was driven by the nom-acc condition, where ratings for sentences without a light verb were the lowest of all the conditions (mean rating = 2.2). We also found two other main effects: First, nom-acc sentences were rated as significantly better than the nom-nom and dat-nom conditions, pooled (means = 4.2 vs 2.7; β = 2, SE = .3, z = 5.7, p <.0001). This indicates that there is an effect of alignment, in the predicted direction: Sentences whose case array is misaligned with the grammatical and thematic accessibility hierarchies (i.e., nom-nom and dat-nom; see again §2) are rated as worse than those that are aligned (e.g., nom-acc). The second main effect showed that dat-nom sentences were also rated as lower than nom-nom sentences (means = 2.6 vs. 2.8; β = .7, SE = .3, z = 2.1, p = .03). Interestingly, sentences with a non-distinct case array (nom-nom) were, in fact, rated better than those whose NP arguments are marked with distinct cases (dat-nom), rather than worse, as might be expected under the distinctness hypothesis. We will return to this finding in our discussion section (§3.3). Lastly, there was a significant interaction of nominative object (i.e., nom-acc vs. nom-nom and dat-nom, pooled) with light verb (β = 4.9, SE = .3, z = 15.9, p <.0001). We explored this interaction by running a second model with nested planned comparisons, asking whether there was an effect of light verb at each sentence type. This model confirmed that ratings to nom-acc sentences were significantly better with a light verb than without (means = 6.1 vs. 2.2; β = 5.8, SE = .55, z = 10.5, p <.0001), but presence of a light verb did not lead to a statistically significant difference in ratings for either nom-nom or dat-nom sentences (ps >.18); Indeed, the interaction of double nominative with light verb was also not significant (p = .5).

3.3. Discussion

The results of this experiment demonstrated, first, that sentences with nominative-marked objects (nom-nom and dat-nom) were consistently rated lower in acceptability compared to sentences with accusative-marked objects (nom-acc). This pattern reflects the influence of both local and global constraints: locally, there is a general dispreference for nominative objects, while globally, certain subject–object case combinations are penalized due to misalignment and lack of case distinctness. Specifically, this finding follows from the alignment hypothesis, which penalizes violations of the alignment of morphological case, according to Otsuka’s (Reference Otsuka, Johns, Massam and Ndayiragije2006) accessibility hierarchy, with grammatical function and thematic role (as in nom-nom and dat-nom constructions). The second important finding was that nom-nom sentences were rated as higher than dat-nom ones, regardless of whether a light verb was present or not. This is especially noteworthy because it is the opposite result from that which is predicted by the distinctness hypothesis: The non-distinct nom-nom construction was the better-rated one of the two. This result suggests that the acceptability of these constructions cannot be explained by distinctness alone. Instead, the overall pattern supports a model in which both local (nom object case marking) and global constraints (case alignment of the entire clause) interact, with cumulative effects on acceptability. Does this mean that sentences with two different-cased NPs (e.g., dat-nom) are necessarily dispreferred over those with the same case (e.g., nom-nom)? We think that this would be too strong of a conclusion to draw, because there is a plausible alternative reason why dat-nom would be worse than nom-nom, owing to multiple misalignments at the global level. As illustrated in (13a), only the object is misaligned in nom-nom constructions (higher nom ➔ lower object theme). But in dat-nom (13b), both the subject (lower dat ➔ higher experiencer subject) and object (higher nom ➔ lower object theme) suffer from misalignment. It could well be that this additional misalignment in dat-nom is what leads to the lower ratings for these sentences.

This observed gradience in acceptability supports a model in which constraint violations are cumulative, as captured by Harmonic Grammar models. Rather than a categorical distinction between a sentence being ‘grammatical’ versus ‘ungrammatical’, our results suggest that the acceptability of a construction is a function of the number (and severity) of different constraints. Any type of misalignment between case, grammatical function, and thematic role incurs a penalty in acceptability, but the magnitude of the penalty depends on the nature and locus of the misalignment. Local violations, such as assigning nominative case to an object/theme, are predicted to incur larger penalties, while compounded global violations (as in dat-nom) add smaller, incremental costs, consistent with cognitive models of diminishing sensitivity (Gibson Reference Gibson, Marantz, Miyashita and O’Neil2000, Wilson & Dillon Reference Wilson and Dillon2022). The observed pattern, where the acceptability drop from nom-acc to nom-nom is larger than from nom-nom to dat-nom, indeed matches with these predictions. With this in mind, we now focus our attention on the alignment hypothesis, which we spell out below, in (14).

We reason that the MTG exists because all three of (i) case morphology, (ii) thematic role, and (iii) grammatical function are different kinds of means to the same end. They typically serve as converging cues that help the language-user infer the event role of an NP in relation to the verb. Although these cues do not always specify the role of an NP, because interpretation also depends on the lexical semantics of the verb and the presence or absence of other arguments. For this reason, the meaning of the sentence is easier to construct when each of these three pieces of information points to the same conclusion about what an NP’s event role is; in other words, this alignment reduces ambiguity and facilitates the assignment of event roles to each NP. It is important to also note that nominative object constructions are overwhelmingly restricted to predicates with experiencer subjects. Thus, the acceptability differences observed here do not reflect a difference in the core thematic roles assigned by the predicates: in all cases, the subject is an experiencer and the object is a theme. While both dative and nominative subjects can be experiencers, dative subjects are typically interpreted as less agentive or less involved in initiating the mental state, consistent with cross-linguistic findings on non-nominative subjects. Thus, the acceptability patterns observed here reflect not just morphological distinctness, but also the pragmatic inferences about agentivity that arise from the combination of case, predicate, and light verb.

While our results are consistent with the predictions of the MTG alignment hypothesis, we acknowledge that alternative explanations are possible. Processing-based theories that link acceptability to expectation or familiarity (e.g., Hale Reference Hale2001, Reference Hale2006, Levy Reference Levy2008) could also predict a preference for more frequent or more canonical argument structures. It may be that the higher acceptability of nom-acc constructions, for example, reflects their greater frequency in Korean, as confirmed by our own corpus findings (see Section 4). Interestingly, both the results of our experiment and the hypothesis in (14) raise an important question: Under what conditions are nom-nom and dat-nom configurations acceptable? We investigate this question next by looking for attested examples in a Korean language corpus.

4.1. Method

The corpus was searched for instances of the Korean psych verbs that were used in the previous experimental design shown in Section 3. This approach allows for a comparative analysis against the low acceptability ratings observed in the experiment. A total of 24 psych verbs were selected; see list in (15). Additionally, 44 distinct verb conjugations across bare-form, ha-form, and ci-form psych verbs were compiled (see Appendix A for details on conjugations). These conjugations represent the most frequently occurring forms in the corpus and were identified through a filtering process that examined verb tense, aspect, and morphological form.

The initial filtering of the Modu Corpus based on these verb conjugations resulted in a dataset of approximately 353,837 sentences. The corpus sentences were then filtered according to three criteria, which allowed for the isolation of transitive uses of psych verbs, with clearly identifiable subjects and objects that could be tagged for their properties (that is, tokens where both the subject and object are marked by a single overt marker: nominative, accusative, dative, or a topic marker. These are laid out in (16).

By applying these criteria, the initial dataset was filtered down to a subset of sentences. A total of 4,927 instances were selected for further analysis. The analysis is confined to objects in a bare noun form or those accompanied by a single determiner within matrix clauses.

4.2. Results and discussion

The distribution of case markings on subjects and objects is summarized in Table 3. Descriptively, we first note that most of the sentences had a subject that was marked by the Korean top(ic) marker -nun (4,556, or 92%). Only 354 (7.2%) had a nom-marked and fewer still had a dat-marked subject – only 17 (0.3%).

Table 3. Counts of sentences with top, nom, and dat subjects, and top, nom, and acc objects, along with proportion of object marking by subject marking (in parentheses). The first of our chi-square analyses (as detailed below) was performed on the raw counts in the bolded square.

We began our analysis by testing whether the distribution of case marking on objects differs depending on whether the clause-mate subject was (i) a topic subject (i.e., a top-nom and top-acc construction), compared with (ii) a nominative subject (i.e., a nom-nom and nom-acc construction). We compared raw frequency counts of these constructions using a Pearson’s chi-square test (see the bold boxing in Table 3). The results revealed a highly significant association between subject and object case marking (χ² = 109.71, p <.0001), indicating that nominative objects are much more likely to occur with topic-marked subjects than with nominative-marked subjects. Proportionally, nominative objects are much more likely to occur with topic-marked subjects (top-nom) than with nominative subjects (nom-nom) (27% vs. 2%). In contrast, accusative objects are less frequent with topic-marked subjects (top-acc) than with nominative subjects (nom-acc) (97% vs. 72%). In short, constructions with nom objects more frequently have top subjects than nom ones. By way of reference, the same is not true for constructions with acc objects, which, in fact, occur with nom subjects more than they do with top subjects. Thus, sentences with nom subjects tend to have acc objects. Sentences with nom objects tend to have top subjects. Examples of token sentences from the corpus with top-nom, nom-nom, top-acc, and nom-acc, respectively, are given in (17).

The overall paucity of nom–nom and dat–nom constructions in the corpus is consistent with the influence of morphosyntactic constraints. Specifically, both local constraints (the general dis-preference for nominative objects) and global constraints (the avoidance of certain subject–object case combinations) likely contribute to their low frequency in this type of naturalistic data. Recall that the MTG alignment hypothesis (14) posits that sentences that have maximal alignment of morphological case marking, thematic roles, and grammatical functions are preferred by language users over sentences with one or more misalignments. In sentences with nominative-cased objects, the assigning of a higher-ranked nominative case to a lower-ranked thematic role like a theme (object) disrupts MTG alignment (as was schematized in (13a and b)). The findings from our corpus survey support this hypothesis from a different angle: there are fewer occurrences of nom objects as compared with acc objects, supporting the idea that constructions with nom objects indeed occur less frequently than those with acc objects.

However, we also observe that when sentences do occur with a nom object, they almost always have a top-marked subject. While morphosyntactic distinctness and the MTG alignment hypothesis predict a general dis-preference for nominative object constructions, they are not entirely absent from natural language use. The prevalence of top-marked subjects – especially in sentences with nom objects – suggests that topic marking plays a crucial role in enabling configurations that would have been disfavored. We consider top as a discourse marker, not a canonical case marker: it does not encode grammatical or thematic role in the same way that structural case does but rather marks information structure (Kuno Reference Kuno1987, Lee & Shimojo Reference Lee and Shimojo2016). Morphologically, it occupies the same post-nominal slot as case markers but is not part of the morphological case hierarchy (as in Otsuka Reference Otsuka, Johns, Massam and Ndayiragije2006, Bobaljik Reference Bobaljik, Harbour, Adger and Béjar2008). As such, when a subject is topic-marked, it is exempt from both the global MTG alignment constraint and the local distinctness constraint: the subject is not counted in the calculation of case alignment, nor does it trigger a distinctness violation with a nom object. In effect, top marking ‘rescues’ the alignment of an otherwise misaligned construction by removing the subject from the domain of case competition, as schematized in (18). This means that, for the purposes of distinctness and MTG alignment, a top-marked subject is effectively invisible: it does not participate in structural case competition, nor does it contribute to violations of case identity with a nominative object. It is crucially different from the situation where two NPs bear different allomorphs of the same case (e.g., -i vs. -ka for nom), which would still count as a violation of distinctness since both are functionally nominative.

This interpretation leads to a prediction: If top marking plays a role in rescuing a potential nom-nom construction from misalignment, then top marking should occur more frequently with nom objects (where it can serve this purpose) compared with acc objects (where it fulfills no such ‘rescue role’). We tested this prediction by looking at the number of sentences with nom objects which had (i) top-marked subjects and (ii) non-top-marked (i.e., nom or dat) subjects, and comparing these with the number of sentences with acc objects with top- vs. non-top-marked subjects. The counts are shown in Table 4: 99% of sentences with nom objects have a top subject, compared to 91% for sentences with ACC objects. To assess this association, we performed a Pearson’s chi-square test on the raw frequency counts (see Table 4, bolded cells). This analysis revealed a highly significant association between object case and topic marking of the subject (χ² = 95.24, p <.0001).

Table 4. Counts of sentences with top and non-top subjects with nom and acc objects, along with proportion of subject marking by object marking (in parentheses). The chi-square analysis was performed on the raw counts in the bolded square.

This distributional pattern supports the graded linking theory (see Section 2.4). As predicted, structural rescue mechanisms such as top marking with nom objects yield near-categorical frequency (99%), reflecting a grammatical necessity for licensing otherwise disfavored configurations. In contrast, optional discourse strategies like top marking with acc objects yield high but more variable frequency (91%). Though both frequencies are high, the observed 8% gap is theoretically meaningful: it signals a qualitative distinction between obligatory grammatical rescue and optional discourse marking (i.e., presence of top). This aligns with our broader claim that the magnitude and categorical nature of frequency effects are determined by the nature and locus of constraint violations, not simply by their statistical significance (e.g., Phillips et al. Reference Phillips, Gaston, Huang, Muller and Goodall2021).

Why does top marking rescue a misalignment? We hypothesize that this is connected to the discourse role of the Korean topic marker. Top indicates that the NP which it marks is pragmatically backgrounded (e.g., Lee & Shimojo Reference Lee and Shimojo2016), meaning that it is old or given information (Kuno Reference Kuno1987). Information that is topic-marked is therefore already familiar in the minds of both conversational participants (i.e., the speaker and the addressee). Therefore, we reason that top functions as a marker of discourse status rather than providing information about what role an NP plays in an event, like nom and dat do. This means that it does not participate in the MTG hierarchy, and as such, top-nom constructions are not misaligned. In short, top does not serve to indicate event role; it marks discourse role. This is consistent with Shin and Verhoeven (Reference Shin, Verhoeven, Johannes, Yoko, Yong-Min, Stavros and Elisabeth2009), whose harmonic alignment framework posits that well-formedness in Korean double case constructions depends on the alignment of multiple hierarchies, including syntactic-based, animacy-based, and pragmatic-based ones.

5.1. Concluding remarks

Overall, our experimental and corpus findings are largely consistent with the generalizations reported in previous studies based on native speaker intuition (Lim Reference Lim1997, Mok Reference Mok2003, Choi Reference Choi2009, Kim et al. Reference Kim, Park, Lee and Choi2018, among others). We extend these findings by providing new quantitative evidence from both experimental and corpus-based methods. The findings from both the acceptability judgment experiment and the corpus analysis are summarized as follows.

Low acceptability and low frequency of nominative objects. Both nom-nom and dat-nom structures demonstrate lower acceptability and frequency compared to nom-acc structures. This indicates that the presence of a nominative object, rather than merely the double nominative case, is the critical factor. We note that the gradient acceptability pattern suggests that unacceptability arises from the cumulative violation of multiple constraints, rather than one single monolithic constraint. Our results are consistent with typological work showing that nominative objects are rare and dispreferred cross-linguistically (Silverstein Reference Silverstein and Dixon1976, Malchukov Reference Malchukov, Amberber and de Hoop2005, Keine Reference Keine2010). However, harmonic alignment is a necessary but of course not sufficient condition for grammaticality in Korean. Predicate-specific licensing requirements (i.e., the presence of the light verb -ha-, here) are essential for licensing accusative case in nom-acc constructions (Jung Reference Jung2002, Reference Jung2003, Yeo Reference Yeo2004, Choi Reference Choi2019, Reference Choi2022).

Local and global constraints. Our analysis supports the view that both local and global constraints together influence the acceptability of nominative object constructions. Local constraints relate to the features of individual constituents, such as the nominative object, whose markedness renders it typologically rare and less acceptable in Korean. Global constraints, on the other hand, pertain to the overall configuration of the clause (e.g., the relation between subject marking and object marking). This necessitates that two clause-mate NPs be as morphologically distinct as possible.

The Morphological-Thematic-Grammatical (MTG) Alignment Hypothesis, which we adopt, operates as a global constraint: it assesses the degree to which morphological case, grammatical function, and thematic role are aligned within a clause. The MTG posits that sentences are more acceptable when there is a maximal alignment between morphological case marking, thematic roles, and grammatical functions. In nom-acc constructions, this alignment is achieved because the subject (a higher-ranked grammatical role) is marked with nominative case (higher-ranked case) and functions as an experiencer (a higher-ranked thematic role), while the object (a lower-ranked grammatical role) is marked with accusative case (lower-ranked case) and functions as a theme (a lower-ranked thematic role). This alignment makes nom-acc constructions more acceptable and more frequently occurring, as evidenced by their high rate of occurrence in corpus data.

In nom-nom constructions, however, alignment is disrupted because the object, which is a lower-ranked thematic role (theme), is marked with a higher-ranked unmarked case (nominative). This misalignment between the thematic role and the morphological case marking reduces the acceptability of nom-nom constructions. Similarly, in dat-nom constructions, there is a twofold misalignment: the subject, which is a higher-ranked thematic role (experiencer), is marked with a lower-ranked marked case (dative), and the object, which is a lower-ranked thematic role (theme), is marked with a higher-ranked unmarked case (nominative). This double misalignment further reduces the acceptability of dat-nom constructions compared to nom-nom constructions.Footnote ⁶

The graded penalty framework we adopt also predicts that the size of acceptability and frequency effects systematically reflects the type and locus of constraint violations: local violations (e.g., nominative objects) incur larger penalties, while compounded global violations (e.g., dative subjects combined with nominative objects) add smaller incremental costs. Similarly, the frequency of topic marking with nominative objects, contrasted with more variable topicalization of accusative objects, reflects a principled distinction between grammatical necessity and pragmatic preference (see, e.g., Phillips et al. Reference Phillips, Gaston, Huang, Muller and Goodall2021).

Topicality as a rescue. In our corpus survey, we found that sentences with nominative objects tend to have topic-marked subjects. We proposed that the presence of a topic marker on the subject can help rescue the MTG alignment in sentences with nominative objects. This is because top indicates that the subject is backgrounded information, meaning that it is already known or given. We think this is because top functions as a marker of information status rather than of grammatical/thematic role. In this line, we argue that the topic marker does not participate in the morphological case hierarchy of the MTG alignment system. Unlike canonical case markers (nom, acc, dat), which encode grammatical and thematic relations, top encodes discourse status. When a subject is topic-marked, it is effectively exempt from the case alignment constraints that would otherwise penalize nom-nom or dat-nom configurations.

5.2. Looking forward: Case stacking

We anticipate that further interesting patterns might emerge when case stacking occurs in nom-object constructions, as in (19), repeated from (8d). In (19), a nom marker is stacked over a dat marker.

Exploring such scenarios in future research ought to shed further new light on how morphological, thematic, and grammatical alignments interact with pragmatic factors such as topicality. Critically, our MTG alignment hypothesis provides a framework for explaining the acceptability of different case configurations, especially in NOC structures like nom-nom and dat-nom. This provides a platform for future research looking at how case stacking work, and especially what the role of information structure might be in potentially helping to rescue misalignments. In closing, we make two predictions about this. First, we predict that instances of case stacking ought to be few in number and/or contextually constrained. This is because nom case and dat case provide different kinds of information about the nature of the experiencer role of the subject (i.e., facilitating a more agentive reading vs. a non-agentive reading, respectively). Our second prediction is that top marking should not generally be stackable over dat or nom marking in constructions in nominative objects. This is because top marking would serve to background a subject that is intended to be simultaneously foregrounded, by virtue of its having nom/dat marking in this environment – a situation that would trigger a clash in discourse functions.