A Dual-Path Analysis of Korean Right-Node Raising: Convergences and Tensions with English ()
1. Introduction
Right-node raising (RNR) is a coordination phenomenon in which a single right-peripheral constituent is interpreted in both conjuncts. Classic English examples demonstrate that RNR shows properties associated with two distinct theoretical mechanisms: ellipsis and multidominance. On the one hand, certain English cases allow morphological mismatches and thus resemble PF-deletion structures; on the other hand, RNR can exhibit conjunct-wide semantic effects, favouring a multidominant pivot (Barros & Vicente, 2011; Valmala, 2013; Belk, Neeleman, & Philip, 2023). These mixed properties have motivated a growing body of work arguing that RNR should not be treated as a uniform phenomenon but instead derives from two syntactically distinct paths. The Korean data examined in this study provide an informative testing ground for such a dual-path view. Korean is typologically different from English in its head-final word order, rich inflectional morphology, productive honorification system, and availability of pro-drop. These properties bear directly on the diagnostics that have been used to distinguish ellipsis-based derivations from multidominance-based ones. In particular, Korean RNR interacts with three empirical domains that have been shown cross-linguistically to be revealing: honorific morphology, plurality-related morphology, and the interpretation of reciprocals and quantifiers. The first domain concerns the honorific suffix -si, whose distribution has been widely discussed in Korean morphosyntax (Sohn, 1999; Kwon & Sturt, 2016; Choi & Harley, 2019). When two conjuncts disagree in honorific features, Korean speakers sometimes nevertheless accept an RNR structure. Such tolerance is naturally captured if one conjunct’s predicate undergoes PF-deletion, as in ellipsis-based RNR. The second domain involves the dependent plural -tul, which can generate cumulative readings when shared across conjuncts. This behaviour aligns with a multidominance analysis, where a single pivot is simultaneously merged into both conjuncts and interpreted conjointly. A similar pattern is observed with reciprocals such as selo and with wide-scope quantifiers, whose cross-conjunct interpretations require a structure in which a constituent syntactically dominates both conjuncts (Belk et al., 2023). Beyond these works, a large literature has argued that what is descriptively called “right-node raising” in fact covers a heterogeneous set of phenomena involving across-the-board movement, extraposition, deletion, and coordination reduction (Ross, 1967; Bošković & Franks, 2000; Bošković, 2004; Chaves, 2014; Larson, 2012).
A central question, therefore, is whether Korean truly replicates the dual-path pattern attested in English or whether typological differences create an alternative alignment of diagnostics. To address this, the analysis adopts a version of Wilder’s (1999) pruning-based PF mapping, which allows a single right-edge output to be generated from either an ellipsis derivation or a multidominance derivation. Under this model, syntax provides either a structure containing parallel material eligible for PF-deletion or a structure containing a multiply dominated pivot, and the PF component removes the offending branch to satisfy linearization requirements. The findings presented here support a cautious dual-path analysis for Korean. Evidence from honorific mismatch favors an ellipsis-based derivation, while dependent plurality, reciprocal interpretation, and quantifier scope argue for multidominance. Constraints on coordination that are independently motivated by respectively and by morphological non-distinctiveness also bear on what counts as a legitimate RNR configuration, further supporting the idea that RNR must be integrated into a broader theory of coordination rather than treated in isolation (Dalrymple & Kehler, 1995; Zaenen & Karttunen, 1984). At the same time, certain data reveal tensions: some honorific paradigms admit alternative explanations via pro-drop, and some predicted contrasts are weaker than in English. These complications suggest that while Korean offers independent support for a dual-path view, the empirical picture is not fully categorical and leaves room for further refinement.
2. Pruning, Diagnostics, and Parsimony
A long-standing observation in the literature is that RNR diagnostics fall into two distinct groups (Barros & Vicente, 2011; Belk et al., 2023). Korean shows the same split: one group behaves as evidence for ellipsis, while another behaves as evidence for multidominance. No single derivation captures both sets of facts without additional stipulations. The pruning-based PF mapping, as proposed by Wilder (1999), is a PF operation that resolves linearization by removing one branch of a structure to satisfy linearization requirements. This operation applies to both ellipsis and multidominance inputs, ensuring proper linear order in both cases. For ellipsis-based structures, the operation deletes one conjunct’s predicate under identity, while for multidominance structures, it removes one of the two branches that dominate the shared pivot. This allows both syntactic structures to converge at the PF interface with a shared right-edge output. The aim of this section is to lay out the empirical basis for this split and to show how both clusters can be unified at PF under a pruning algorithm in the sense of Wilder (1999).
2.1. The Ellipsis-Type Cluster: Honorific Mismatches
Korean honorific agreement is clause-bounded: the verbal suffix -si must agree with a local nominative controller within its own clause (Sohn, 1999; Kwon & Sturt, 2016; Jou, 2024). Under RNR, however, -si may surface in a position where only one conjunct provides an appropriate controller. This creates the illusion that -si is agreeing across conjuncts, but the asymmetry in grammaticality shows that the effect arises from PF deletion in the first conjunct. The contrast in (1) illustrates the point:
(1) a. Mary-ka sakwa-lul, kuliko eomeonim-kkeyse banana-lul, sa-si-ess-ta.
“Mary (bought) apples, and Mother bought bananas.”
b.*eomeonim-kkeyse banana-lul, kuliko Mary-ka sakwa-lul, sa-si-ess-ta.
Intended: “Mother (bought) bananas, and Mary bought apples.”
In (1a), -si is locally licensed by eomeonim “mother” in the second conjunct, while the predicate in the first conjunct is deleted. In (1b), deleting the second conjunct’s predicate would strand -si adjacent to the non-honorific subject Mary, which results in a violation. The asymmetry shows that the honorific effect is computed per conjunct, supporting an ellipsis-based derivation.
2.2. Multidominance-Type Diagnostics
Other diagnostics point toward a very different structure—one where the pivot is syntactically shared. These include cumulative agreement with the dependent plural -tul, as well as cross-coordinate dependencies involving reciprocals and wide-scope quantifiers. These effects cannot be reproduced under an ellipsis analysis alone. The clitic -tul usually marks clause-internal subject plurality. Under RNR, however, it may appear even when each conjunct contains only a singular DP.
(2) John-un nonmwun-ul, Mary-nun chayk-ul, yelsimhi-tul ilk-ess-ta.
“John read articles, and Mary read books diligently.”
The plural marking on yelsimhi-tul does not reflect morphological properties of either subject individually but rather the cumulative event formed by the conjunction. This requires a shared syntactic domain—naturally captured if the pivot is multidominated by both conjuncts. Parallel effects arise with reciprocals (selo) and wide-scope quantifiers such as modeun…-eul. When such elements occur inside the pivot, their interpretive domain may extend across both conjuncts. This is unexpected under ellipsis, which would necessarily reconstruct separate QPs in each conjunct, but is predicted under multidominance.
2.3. Interaction and the PIE ≫ PIM Asymmetry
The two groups of diagnostics are not freely combinable. Korean reveals a directional restriction: properties indicative of ellipsis (PIE), such as honorific mismatch, may precede properties indicative of multidominance (PIM), such as cumulative -tul, but the reverse order is excluded.
(3) PIE ≫ PIM (grammatical)
John-un sakwa-lul, eomeoni-kkeyse-nun pwue-lul,
kyelkuk sa-si-ko mek-tul-ess-ta.
“John bought apples, and mother bought pears, and they finally ate them.”
(4) *PIM ≫ PIE (ungrammatical)
John-un sakwa-lul, eomeoni-kkeyse-nun pwue-lul,
mek-tul-ko sa-si-ess-ta.
Intended: “John ate apples and mother ate pears, and they bought them.”
The contrast suggests that multidominance-based effects (PIM) must be fixed in the syntax before any ellipsis-driven honorific mismatch (PIE) is introduced. This directional compatibility is summarized in Table 1:
Table 1. Split diagnostic profile of Korean RNR.
Diagnostic Type |
Property |
Ellipsis (RNR-E) |
Multidominance (RNR-MD) |
PIE |
Honorific mismatch |
✔ |
✗ |
PIM |
Cumulative -tul |
✗ |
✔ |
|
Reciprocals/wide scope |
✗ |
✔ |
2.4. Pruning and Parsimony
The architecture assumed here is compatible with a broadly minimalist view in which narrow syntax proceeds in phase-based cycles, while many morphophonological operations apply at the PF branch of the derivation (Chomsky, 1995, 2001, 2008). On this view, vocabulary insertion and agreement are late, post-syntactic processes in the sense of Distributed Morphology and related work (Halle & Marantz, 1993; Embick & Noyer, 2001; Bobaljik, 2008), and pruning can be seen as one among several PF operations that manipulate structure once syntax has done its job. Rather than deriving all diagnostics from one operation, a simpler architecture emerges if syntax freely generates either an ellipsis input or a multidominance input, while PF linearization applies a single pruning algorithm (Wilder, 1999) to avoid No-Tangling violations.
Under an ellipsis input, pruning deletes one conjunct’s rightmost VP under identity, producing the pivot. Under a multidominance input, pruning removes one of the two branches pointing to the shared pivot to permit linearization. The PF procedure is therefore uniform, while the syntactic inputs remain distinct—predicting the split diagnostic profile observed in Korean.
3. The Multidominance Cluster: Cumulative Agreement, Reciprocals, and Wide Scope
A second major cluster of diagnostics for Korean RNR supports a multidominance analysis. Unlike the honorific mismatch pattern, which can be derived through PF-deletion in only one conjunct, the diagnostics in this section require a shared syntactic domain across conjuncts. Korean provides three robust empirical sources for this conclusion: (i) cumulative agreement triggered by the dependent plural -tul; (ii) reciprocal interpretation with selo; and (iii) wide-scope readings of quantificational phrases. In all three cases, the interpretive domain extends across both conjuncts, a pattern incompatible with purely ellipsis-based RNR.
3.1. Cumulative Plurality and the Dependent Plural -tul
The Korean dependent plural -tul typically marks the plurality of the subject or object within a single clause. It normally scopes over the event described by a single predicate and does not induce distributive or cumulative readings unless supported by the syntactic structure. Under RNR, however, -tul may appear in the pivot even when each conjunct contains only a singular DP, yielding a reading that aggregates participants or events from both conjuncts.
(5) John-un nonmwun-ul, Mary-nun chayk-ul, yelsimhi-tul ilk-ess-ta.
“John read articles, and Mary read books diligently.”
In (5), neither John nor Mary triggers plural marking individually, yet yelsimhi-tul receives a cumulative interpretation based on the joint event formed by the coordination. Under an ellipsis-based analysis, each conjunct would be syntactically independent and therefore unable to support an interpretation in which -tul ranges over a set of events derived from both conjuncts. The cumulative effect instead follows naturally if the pivot is multiply dominated by both conjuncts, creating a domain in which plural morphology can be interpreted collectively. The cumulative interpretations triggered by -tul fit naturally into standard accounts of plurality as involving sums, covers, and event plurality (Partee, ter Meulen, & Wall, 1993; Schwarzschild, 1996; Lasersohn, 1995). In particular, the fact that -tul can take a cumulative reading over the two conjuncts resonates with Brisson’s (2003) observation that plural morphology and modifiers like all can be sensitive to contextually salient pluralities rather than to the morphosyntactic features of a single argument.
The same logic holds for cases where -tul appears on the object or adverbial inside the pivot. The plural morphology does not track features of a single argument but rather the semantic contribution of the entire coordination. This behavior aligns strongly with the predictions of multidominance approaches (Barros & Vicente, 2011; Belk et al., 2023), under which a single pivot node is shared across both conjuncts.
3.2. Reciprocals (Selo) across Conjuncts
A second line of evidence arises from reciprocals such as selo, whose interpretation requires a domain containing at least two participants. The behaviour of selo is also consistent with recent work on the diachronic and pragmatic development of Korean reciprocals, which shows a shift from purely reciprocal to more competitive and subjectified readings (Jendraschek & Koo, 2023). The fact that selo can still support a genuinely reciprocal interpretation across conjuncts in RNR contexts suggests that the syntactic configuration can override some of these pragmatic tendencies. Korean reciprocals are clause-bound: they cannot take antecedents across independent clauses or across unrelated predicates. Under RNR, however, selo inside the pivot can take its antecedents jointly from the subjects of both conjuncts.
(6) John-kwa Mary-nun chayk-ul, Tom-kwa Ann-un nonmwun-ul, selo ilk-ess-ta.
“John and Mary read books, and Tom and Ann read articles; they read them together.”
Here selo receives a cross-conjunct reciprocal reading. Under an ellipsis derivation, each conjunct would contain its own local copy of the predicate, and selo reconstructed in each conjunct would only be able to take the local subject pair as its antecedent. The attested interpretation—one in which all four individuals jointly participate in a reciprocal action—is incompatible with a conjunct-local structure. It instead requires a single syntactic constituent that is accessible from both conjuncts, consistent with a multidominant pivot.
3.3. Wide-Scope Quantifiers and Cross-Conjunct Interpretation
Wide-scope quantificational phrases such as modeun…-eul (“every…-Acc”) further support the multidominance cluster. Korean QPs exhibit well-known scope constraints and generally cannot take scope outside their local clause unless embedded in a structure that syntactically dominates the relevant domain.
Under RNR, however, certain QPs inside the pivot can take wide scope across both conjuncts. This is illustrated by quantifiers inside right-peripheral objects or adverbials, which yield interpretations that quantify over the events contributed by both conjuncts.
(7) John-un sakwa-lul, Mary-nun phwul-lul,
modeun salam-i mek-ess-ta.
“John ate apples, and Mary ate grass; everyone ate them.”
Again, such a reading is unavailable under an ellipsis analysis, where each conjunct would contain its own reconstructed quantifier phrase. Only multidominance allows the quantifier to originate in a structural position that simultaneously outscopes both conjuncts.
Together, cumulative plurality, reciprocal interpretation, and wide-scope quantification form a coherent multidominance cluster in Korean RNR. All three diagnostics require a pivot that is syntactically merged into both conjuncts simultaneously. Under an ellipsis-based derivation, these effects cannot arise: ellipsis deletes one conjunct’s predicate and reconstructs structure within each conjunct, thereby preserving the locality of interpretation. The multidominance effects therefore establish the necessity of a second syntactic path for Korean RNR.
4. Honorific Mismatch, Ellipsis, and Hybrid Pivots
Honorific morphology in Korean provides one of the clearest diagnostics for ellipsis-based RNR. The verbal suffix -si is canonically clause-bound and must be licensed by a local nominative subject with [+HON] features (Sohn, 1999; Kwon & Sturt, 2016; Choi & Harley, 2019). Because the licensing relationship is strictly local, -si cannot normally appear on a predicate whose subject lacks honorification. Korean RNR, however, produces apparent violations of this restriction. These violations are not genuine long-distance agreement; instead, they reflect the interaction of local licensing with PF deletion in one conjunct. This section examines the properties of -si, demonstrates how honorific mismatch diagnoses ellipsis-based RNR, and shows how the diagnostic interacts with multidominance under hybrid pivots. “Hybrid pivots” refer to structures that simultaneously exhibit properties of both ellipsis and multidominance. In such structures, one conjunct’s material undergoes PF deletion (as in ellipsis), while the pivot is syntactically shared across both conjuncts (as in multidominance). This dual property allows these pivots to show diagnostics from both ellipsis-type and multidominance-type phenomena.
4.1. Local Licensing of -si and Apparent Cross-Conjunct Agreement
A key property of -si is that it cannot be licensed across clause boundaries or by discourse-level topics. Its distribution is fully determined within the minimal clause containing the predicate (Sohn, 1999). In standard monoclausal structures, substituting a non-honorific subject for an honorific one immediately disrupts the agreement pattern. Korean’s rigid honorific system, therefore, provides a strong test for identifying where predicate licensing actually occurs.
Under RNR, one conjunct may contain an honorific subject while the other does not. When the shared pivot contains -si, apparent cross-conjunct licensing arises: -si surfaces in the final position even though only one conjunct supplies a compatible controller. The interaction between honorification and other dependency types in Korean has been extensively studied in work on multiple subject constructions and long-distance anaphora, which show that both Agree-based dependencies and discourse-driven interpretations are available in the language (Kim & Sells, 2007; Kim & Yoon, 2009, 2020; Han, Storoshenko, Leung, & Kim, 2015). These results highlight that apparent long-distance effects in Korean often conceal more local structural licensing, a point that aligns well with the present use of -si as a clause-bounded diagnostic. The contrast in (8) illustrates this pattern, reproduced from the general form already discussed earlier:
(8) a. Mary-ka sakwa-lul, kuliko eomeonim-kkeyse banana-lul, sa-si-ess-ta.
“Mary (bought) apples, and Mother bought bananas.”
b. *eomeonim-kkeyse banana-lul, kuliko Mary-ka sakwa-lul, sa-si-ess-ta.
Intended: “Mother bought bananas, and Mary bought apples.”
The grammaticality of (8a) arises because -si is locally licensed by eomeonim in the second conjunct, while the first conjunct’s predicate is deleted. In (8b), deletion would leave -si adjacent to the non-honorific subject Mary, violating local agreement. Because the ungrammaticality follows from the requirements on local licensing rather than from properties of coordination per se, this asymmetry strongly favors an ellipsis-based analysis (Jou, 2024). Under a multidominance analysis, by contrast, nothing forces -si to be licensed by only one conjunct, and the predicted availability of “shared licensing” incorrectly predicts (8b) to be acceptable. Honorific mismatch therefore provides a structural argument against a pure multidominance account for these cases.
4.2. Hybrid Pivots and the PIE ≫ PIM Restriction
Honorific mismatch is an ellipsis-type diagnostic (PIE). Multidominance-type diagnostics (PIM) include cumulative -tul, cross-conjunct reciprocals, and wide-scope quantifiers. When these diagnostics combine, Korean reveals a striking directional restriction: PIE material may precede PIM material, but PIM may not precede PIE. As shown earlier, this asymmetry is illustrated by contrasts such as:
(9) PIE ≫ PIM
John-un sakwa-lul, eomeoni-kkeyse-nun pwue-lul,
kyelkuk sa-si-ko mek-tul-ess-ta.
“John bought apples and Mother bought pears, and they finally ate them.”
(10) *PIM ≫ PIE
John-un sakwa-lul, eomeoni-kkeyse-nun pwue-lul,
mek-tul-ko sa-si-ess-ta.
Intended: “John ate apples and mother ate pears, and they bought them.”
The difference cannot be explained by a simple morphological filter or an independent constraint on ordering. Instead, the restriction follows from the architecture of the derivation. Multidominance relations (PIM) must be established in the core syntax, while ellipsis-driven properties (PIE), such as honorific mismatch, are resolved later at the PF interface, after a single overt conjunct has been selected. Multidominance structures require both conjuncts to contribute material to a single pivot before PF pruning removes one of the two branches (Wilder, 1999; Barros & Vicente, 2011; Belk et al., 2023). Ellipsis-based RNR, on the other hand, requires that the conjunct containing the sensitive material (-si) is the one that survives pruning. Hybrid pivots thus impose a dependency on derivational timing:
PIM material must be established in the syntax prior to pruning.
PIE material must be introduced only after pruning has determined which conjunct contributes the overt pivot.
This sequential dependency predicts exactly the observed asymmetry. When multidominance morphology (-tul, reciprocals, wide-scope QPs) precedes honorific mismatch, pruning forces the deletion of the wrong conjunct, stranding mismatched -si—yielding ungrammaticality.
4.3. Alternative Analyses and Points of Tension
While honorific mismatch patterns support an ellipsis-based path, some Korean data create interpretive ambiguities. For example, certain examples that superficially appear to force an RNR derivation may alternatively be parsed as biclausal structures with pro-drop in the initial clause. In these cases, pro-drop allows agreement to be satisfied without an overt subject in the initial conjunct. This analysis, however, does not account for the systematic nature of the honorific mismatch observed in the data, where the mismatch is consistently linked to PF-deletion in one conjunct. On the other hand, the ellipsis-based RNR analysis more directly captures the pattern by showing that the honorific mismatch arises when one conjunct undergoes PF-deletion, leaving the other conjunct’s subject to license -si, without relying on independent clause structures or pro-drop phenomena. This possibility reduces the diagnostic force of -si in some contexts, since pro can satisfy agreement conditions without overt subjects. Similarly, some examples involving multiple occurrences of -si or multiple possible controllers produce weaker or more variable intuitions than their English counterparts. Several multidominance-based diagnostics also show variability: wide-scope QP readings are sometimes degraded for certain speakers unless provided with supporting context, and reciprocal readings can shift between pairwise and group interpretations depending on discourse factors. These uncertainties do not undermine the dual-path architecture, but they caution against overly categorical claims.
Together, these points mirror observations in previous studies that RNR diagnostics can be fragile and that some effects depend on subtle interactions between syntax, prosody, and discourse (Belk et al., 2023). Korean contributes further nuance: its rich agreement and honorification system allow more fine-grained diagnostics, but also introduce new potential confounds such as pro-drop and independent-clause reinterpretation.
5. Variation, Constraints, and Theoretical Implications
The Korean data presented above show that right-node raising (RNR) cannot be captured by a single uniform mechanism. Instead, the language exhibits a dual-path architecture in which both ellipsis-based and multidominance-based derivations are independently attested. While the preceding sections have established the empirical motivations for this split, several broader issues require attention. These concern cross-linguistic variation, the interaction of diagnostics, limitations of the evidence, and implications for the syntax-PF interface. This section expands on each of these themes and identifies open questions that follow from the Korean pattern.
5.1. Cross-Linguistic Variation: Korean vs. English
A first issue concerns how the Korean pattern fits into the broader typology of RNR across languages. English has long been known to display both ellipsis-type and multidominance-type properties (Barros & Vicente, 2011; Belk, Neeleman, & Philip, 2023). However, the relevant diagnostics are often indirect, because English has little inflectional morphology and relatively flexible word order. Korean sits at a very different point in the typological space: it combines strict head-final order with rich verbal honorification and dedicated plural morphology. This configuration makes it possible to isolate diagnostics that would be much harder to detect in English. The comparison suggests that the dual-path architecture itself is not uniquely Korean. Rather, what is special about Korean is the visibility of the two paths. The honorific suffix -si and the dependent plural -tul overtly mark features that, in English, are either not morphologically realized or only indirectly detectable. Work on cyclic linearization and Korean coordination already shows that head-final syntax interacts closely with PF mapping (Ko, 2007; Park, 2009), and the Korean RNR facts can be seen as a particularly clear instance of this interaction. In languages where agreement and plurality are less morphologized, similar underlying structures may exist, but their diagnostics are correspondingly weaker or more easily confounded.
Cross-linguistic research further supports a parameterized view of how RNR is realized at the interface. Belk et al. (2023) propose that languages differ in how tolerant they are of feature conflict at PF and in how readily they resort to multidominance for coordination. Evidence from languages with rich case and agreement morphology, such as the Finnish data discussed by Bjorkman (2021), shows that syncretism can sometimes disguise feature mismatches in a way reminiscent of Korean -si and -tul, allowing a single pivot to satisfy the requirements of both conjuncts. By contrast, in languages like French, RNR appears to be much more tightly constrained by recoverability and morphological compatibility, encouraging analyses that are closer to pure ellipsis (Abels, 2004). Greek provides an additional point of comparison: rich inflectional paradigms and pervasive syncretism create conditions under which a single right-edge DP can in principle satisfy multiple case requirements, but descriptive grammars emphasize that speakers often avoid such configurations in favour of repetition or clausal coordination (Smyth, 1956; Holton, Mackridge, Philippaki-Warburton, & Spyropoulos, 2012; Holton, Mackridge & Philippaki-Warburton, 2016). This again suggests that PF and morphology place language-specific bounds on how far multidominance-type solutions can be exploited. Within this typology, Korean emerges as a PF-rich/MD-rich system, where both paths are active and where the interaction of honorification, plurality, and pruning yields particularly sharp diagnostics.
5.2. Variation in Diagnostic Strength and Speaker Judgments
A second theme is the variability in how strongly individual diagnostics are perceived by speakers. Even within Korean, not all speakers respond identically to the same examples. As in the thesis, the judgments reported here are based on a small set of primary consultants, with additional speakers consulted for key contrasts. For these speakers, the honorific mismatch paradigm is consistently robust: the contrast between examples like (1)/(8a) and (1b)/(8b) is categorical, and the directionality of -si licensing is clear. This is unsurprising given independent evidence that honorific concord is strictly local to the clause containing the finite verb (Sohn, 1999; Choi & Harley, 2019; Kwon & Sturt, 2016). By contrast, multidominance diagnostics such as cumulative -tul, cross-conjunct selo, and wide-scope modeun…-eul can be more gradient. Some speakers accept cumulative or wide-scope readings only in carefully constructed contexts, and even then they sometimes report a preference for interpretations that minimize cross-conjunct dependencies. This gradience is consistent with general observations about the subtlety of scope and plurality judgments (Lasersohn, 1995) and about the difficulty of isolating semantic from processing factors in complex sentences (Madigan, 2015; Merchant, 2001). For instance, the cognitive load of keeping track of multiple participants across conjuncts may make reciprocal readings less accessible unless the discourse strongly supports a joint-event interpretation. The variability does not undermine the dual-path proposal, but it affects how the diagnostics should be used. Rather than relying on single, isolated examples, the argument for multidominance has to be built from converging patterns: cumulative agreement, reciprocal binding, and wide scope all point in the same direction, even if any one test on its own is somewhat soft. Likewise, the PIE ≫ PIM asymmetry becomes particularly compelling when it is observed across multiple paradigms—-si plus -tul, -si plus selo, -si plus modeun…-eul—rather than in a single minimal pair. This strategy mirrors the methodological caution recommended in other work on RNR, where combinations of diagnostics are used to compensate for the fragility of individual tests (Belk et al., 2023).
5.3. Hybrid Pivots and Derivational Timing
A central theoretical consequence of the Korean pattern lies in the behaviour of hybrid pivots—structures that simultaneously exhibit PIE-type and PIM-type properties. The existence of such pivots shows that ellipsis and multidominance are not mutually exclusive options that compete in a global fashion. Instead, they can cooperate within a single derivation, with one path embedded inside the other. Korean suggests a specific asymmetry: an ellipsis-derived pivot (RNR-E) can host a right-peripheral sub-pivot created by multidominance (RNR-MD), but not vice versa. On the account adopted here, this asymmetry falls out from the interaction of syntax and PF pruning (Wilder, 1999; Bachrach & Katzir, 2009). Syntax first builds one of two possible inputs: either (i) parallel conjuncts whose right edges are eligible for PF deletion, or (ii) a structure in which two T heads share a single pivot via multidominance. PF then applies a single pruning algorithm that deletes one of two conflicting branches at the right edge to avoid No-Tangling violations. If syntax has introduced multidominance first, pruning can safely remove the redundant branch while preserving the shared pivot, yielding an RNR-MD configuration that supports PIM diagnostics such as cumulative -tul and cross-conjunct selo. Ellipsis-like behaviour can then arise inside this pivot if additional, purely phonological deletion of material is licensed.
The reverse ordering, however, is problematic. If ellipsis is resolved first in such a way that -si is stranded with a non-honorific subject, PF pruning has no way to rescue the violation without destroying the multidominant pivot. This is exactly what is seen in the ungrammatical PIM ≫ PIE configuration in (10). In other words, hybrid pivots reveal that the dual-path architecture is derivationally ordered: multidominance effects must be fixed early, at the level where two conjuncts share a syntactic pivot, whereas ellipsis-type effects can arise only once the PF algorithm has determined which conjunct contributes the overt right-edge material. The Korean data therefore support a view of RNR in which surface linearization is the outcome of an interaction between syntactic options and an economy-driven PF procedure, rather than a single uniform transformation (Barros & Vicente, 2011; Belk et al., 2023).
5.4. Limitations and Directions for Future Research
Despite the advantages of Korean as a testing ground, the present analysis faces several limitations that future work should address. First, the empirical base is still relatively narrow. Many examples used here are carefully constructed to isolate particular diagnostics, and although they have been checked with multiple speakers, they do not yet constitute a systematic corpus of naturally occurring RNR. Such methods are necessary because complex syntactic configurations, like those involving ellipsis and multidominance in RNR, rarely appear in natural language corpora. Constructed examples and consultant judgments allow us to test the grammaticality of these rare configurations and provide insights into syntactic phenomena that are difficult to capture in naturally occurring data. A more comprehensive picture would emerge from combining constructed examples with corpus searches in large Korean text collections and from experimental work that measures graded acceptability and processing difficulty for PIE-, PIM- and hybrid structures (Kwon & Sturt, 2016). Second, the analysis has remained largely silent on prosody, even though PF pruning is, by hypothesis, an operation at the syntax–phonology interface. If pruning is genuinely a PF phenomenon, one might expect PIE-only, PIM-only and hybrid pivots to show different prosodic profiles—for instance, in terms of phrasing breaks or prominence on the pivot. Detailed phonetic and phonological studies would be necessary to determine whether such prosodic signatures exist and how they interact with native speakers’ intuitions about RNR (Bachrach & Katzir, 2007, 2009). Third, the discussion has focused almost exclusively on RNR, leaving aside related phenomena such as Korean left-node raising (LNR). Recent work on LNR in Korean suggests that case mismatches and dependent plural markers behave differently from their RNR counterparts (Kim, Kim, & Jung, 2020; Kim, Choi, & Lee, 2023), raising the question of whether the dual-path architecture extends to left-edge coordination. If LNR systematically fails to show a PIE/PIM split, that would support the idea that the dual-path model is specific to right-edge configurations; if, however, LNR also reveals independent ellipsis- and multidominance-type diagnostics, the scope of the pruning-based account may need to be broadened. Finally, more work is needed on variation across dialects and related languages. As already noted in the thesis, Jejueo differs from Standard Korean in both its plural marking and its honorification system (O’Grady, Yang, & Yang, 2019). The diagnostics developed here rely on -si and -tul; it is therefore an open question whether analogous cues can be found in Jejueo or in other honorific languages such as Japanese. Cross-linguistic comparison along these lines would test whether the three parameters implicitly at work in the Korean analysis—honorific tolerance, visibility of plurality in morphology, and PF pruning defaults—are sufficient to capture the range of RNR systems observed in the typology sketched by Belk et al. (2023).
To summarise, the Korean data do not merely add another language to the list of RNR systems: they sharpen the empirical basis for a dual-path architecture in which ellipsis and multidominance co-exist and interact in constrained ways. The PIE/PIM split, the asymmetry in hybrid pivots, and the role of PF pruning together suggest that RNR is best analysed as the convergence of two derivations under a single interface operation. At the same time, variation across languages, dialects, and speakers, and the methodological challenges involved in testing subtle diagnostics, all point to the need for further empirical and theoretical work. The picture that emerges is one in which syntax generates multiple options and the interface chooses among them in the most economical way—a view that is entirely in line with broader interface-based approaches to ellipsis and coordination, but which finds particularly clear support in Korean.
6. Conclusion
This paper has argued that Korean right-node raising (RNR) provides strong evidence for a dual-path architecture in which ellipsis-based and multidominance-based derivations coexist. The empirical picture drawn from honorific mismatch, dependent plurality, reciprocal interpretation, and wide-scope quantification shows that no single derivational mechanism captures the full range of observed patterns. Instead, Korean exhibits a systematic split between diagnostics associated with PF-deletion and those requiring a syntactically shared pivot, corroborating a view already proposed for English but made more transparent by Korean’s rich morphosyntax (Sohn, 1999; Kwon & Sturt, 2016; Barros & Vicente, 2011; Belk et al., 2023). Honorific mismatch, driven by the clause-bounded licensing properties of -si, demonstrates that Korean speakers allow RNR structures in which only one conjunct supplies the features necessary for verbal honorification. This asymmetry is best explained by ellipsis-based RNR, where PF deletes the predicate of the conjunct lacking a compatible controller. By contrast, cumulative -tul marking, cross-conjunct reciprocals (selo), and wide-scope quantifiers show that Korean RNR can also create a domain in which interpretive dependencies extend across both conjuncts, a pattern consistent only with multidominance. A notable contribution of the Korean pattern is the behaviour of hybrid pivots, where both classes of diagnostics appear in a single right-peripheral constituent. The directional restriction—PIE ≫ PIM but not PIM ≫ PIE—supports the conclusion that multidominance must be established in syntax prior to PF pruning, while ellipsis-driven honorific mismatch can arise only after pruning determines which conjunct contributes the overt pivot. This derivational sequencing aligns naturally with pruning-based PF models (Wilder, 1999; Bachrach & Katzir, 2009) and further motivates a two-input, one-output view of RNR.
At the same time, several aspects of the evidence remain open for future investigation. Speaker variation in PIM diagnostics, potential prosodic correlates of pruning, and comparisons with related phenomena such as Korean left-node raising all point to avenues for further study. Cross-linguistic work—examining languages with different honorification, plurality, and agreement systems—will be essential for determining the extent to which the Korean pattern reflects universal properties of the syntax-PF interface, rather than language-specific morphology. Overall, Korean RNR offers unusually clear evidence that right-edge convergence can arise through more than one syntactic pathway, unified only at PF. The PIE/PIM split and the hybrid pivot asymmetry in Korean play a key role in refining the theoretical debate on whether RNR is a uniform or heterogeneous phenomenon in universal grammar. These patterns demonstrate that RNR cannot be captured by a single derivational mechanism, but instead involves distinct syntactic paths (ellipsis-based and multidominance-based). The Korean data, therefore, strengthen the broader claim that coordination structures are best analyzed not through a single uniform derivation, but through a system in which syntax provides alternative inputs, with the PF component selecting among them based on economy and linearization.