Highlights
-
• Native speakers associate the subject honorific suffix with honorifiable subjects only
-
• L2-Korean learners perceive and apply this suffix as a general politeness marker
-
• Semantic cues overshadow morphosyntactic agreement in learners’ honorific use
-
• Low cue validity, functional overlap and socio–pragmatic complexity hinder acquisition
1. Introduction
One notable feature of second/non-dominant language (L2) knowledge is its comparatively noisier representations than those of first/dominant language (L1) knowledge in a learner’s mind (Futrell & Gibson, Reference Futrell and Gibson2017; Tachihara & Goldberg, Reference Tachihara and Goldberg2020). This noisiness, marked by vague and less specific linguistic formulations, stems from weaker memory traces for L2 compared to L1. Contributing factors include language use/learning experience (Ellis, Reference Ellis2002; Shin et al., Reference Shin, Jung and Yang2025), cross-linguistic influence (Frenck-Mestre et al., Reference Frenck-Mestre, Kim, Choo, Ghio, Herschensohn and Koh2019; Jackson & Dussias, Reference Jackson and Dussias2009), cognitive load (Cunnings, Reference Cunnings2017; Pozzan & Trueswell, Reference Pozzan and Trueswell2016), and individual profiles (Dąbrowska & Street, Reference Dąbrowska and Street2006; Linck & Weiss, Reference Linck and Weiss2015). Consequently, learners exhibit a reduced capacity to utilize target knowledge in language tasks (Grüter & Rohde, Reference Grüter and Rohde2021; Robenalt & Goldberg, Reference Robenalt and Goldberg2016), resulting in learning trajectories that diverge from those of L1 acquisition (Dąbrowska et al., Reference Dąbrowska, Becker and Miorelli2020; Slabakova, Reference Slabakova2014).
Extant research on bilingualism and L2 acquisition grounded in this concept has focused on a narrow set of languages – primarily English – producing an English-centric perspective that constrains broader theoretical insights (cf. Blasi et al., Reference Blasi, Henrich, Adamou, Kemmerer and Majid2022). Expanding the typological range is therefore essential to refine acquisition models and reveal language-specific adaptations rarely visible in dominant languages. Korean, although increasingly popular as an L2 target, remains underexplored in research on L2 knowledge representations (e.g., Frenck-Mestre et al., Reference Frenck-Mestre, Kim, Choo, Ghio, Herschensohn and Koh2019; Kim et al., Reference Kim, Jung and Lee2023; Lee et al., Reference Lee, Shin and Jung2024; Park & Kim, Reference Park and Kim2023). As an agglutinative, Subject−Object−Verb language, Korean employs case particles and verbal morphology to encode grammatical information and permits omission of contextually recoverable elements (Sohn, Reference Sohn1999). Its extensive honorification system, still underexamined in L2-acquisition research (cf. Kim et al., Reference Kim, Kim and Kim2025), offers a valuable paradigm for examining how learners acquire grammatical functions that encode the speaker’s stance toward discourse referents and how grammatical and socio-cultural constraints are integrated for appropriate use. The present study addresses this gap by examining how learners acquire functions that are structurally governed yet socially grounded. Specifically, we examine how English-speaking learners navigate Korean subject honorification, focusing on the interplay of structural, semantic, and pragmatic cues that shape L2 acquisition trajectories (Brown, Reference Brown, Brown and Yeon2015; Sohn, Reference Sohn1999; see Section 1.2).
1.1. Learning mechanism: cue competition and weighting in L2 acquisition
Given the noisiness of L2 knowledge, successful integration of (non-)linguistic information is essential for achieving target-like competence. During this process, cues from diverse sources compete, with some preferred over others. The Competition Model (Bates & MacWhinney, Reference Bates, MacWhinney, Wanner and Gleitman1982) and its extension, the Unified Model (MacWhinney, Reference MacWhinney, Robinson and Ellis2008, Reference MacWhinney, Gass and Mackey2013), provide a conceptual basis for understanding how competing cues operate in L2 development. The model assumes that learning trajectories are largely shaped by cue strength, which is determined by cue validity as a function of reliability (the consistency and predictiveness with which a cue signals a linguistic outcome) and availability (the ease with which a cue can be accessed or used during language use). It further predicts variation in learners’ performance as a function of task demands, distinguishing offline and online tasks. Learners deploy cues differently across these task types: offline tasks allow reflection on the cumulative validity of relevant cues, whereas online tasks constrain such reflection because of higher cue costs and working-memory limits.
This framework has been influential in elucidating L2 acquisition pathways (e.g., Grüter et al., Reference Grüter, Lau and Ling2020; Tuninetti et al., Reference Tuninetti, Warren and Tokowicz2015; Zhao & Fan, Reference Zhao and Fan2021). Empirical findings consistently show that L2 learners often prioritize non-structural (e.g., semantic) cues over structural (e.g., morphosyntactic) cues. For example, in a visual-world eye-tracking study, Grüter et al. (Reference Grüter, Lau and Ling2020) showed that L2-Chinese learners experienced greater interference from semantically related but grammatically incompatible nouns when processing nominal classifiers than L1 speakers. Thus, whereas L1 speakers are sensitive to both structural and semantic cues, L2 learners rely predominantly on semantic cues in classifier processing tasks. The authors further attribute this divergence to learners treating semantic features as a more reliable (albeit imperfect) cue than strictly grammatical form-class membership.
This view accords with Gass (Reference Gass1986): early-stage L2 learners privilege semantic over syntactic information, with a gradual shift toward greater reliance on syntax as proficiency increases. Deniz (Reference Deniz2022) further reports differential cue weighting across task types: L1 speakers tend to favor syntactic cues in real-time parsing (e.g., self-paced reading) and semantic cues in end-of-sentence tasks (e.g., read-aloud and pen-and-paper production), whereas L2 learners rely on semantic cues across both task types. Clahsen and Felser’s (Reference Clahsen and Felser2006) Shallow Structure Hypothesis likewise addresses cue weighting, proposing that although core parsing principles are shared across L1 and L2 speakers, grammatical constraints are less robust in L2 processing. Consequently, L2 learners are more sensitive to semantic, pragmatic, and other non-syntactic information during sentence processing. Importantly, the account does not claim that L2 parsing is exclusively shallow; with multiple processing streams available, the L2 processor may underuse – or selectively attend to – syntactic information in conjunction with contextual and learner factors during online processing (Clahsen & Felser, Reference Clahsen and Felser2018; Lim & Christianson, Reference Lim and Christianson2013; Omaki & Schulz, Reference Omaki and Schulz2011).
The overshadowing-and-blocking account (Ellis, Reference Ellis2006, Reference Ellis, Robinson and Ellis2008) formalizes how asymmetric weighting between semantic and structural cues shapes L2 acquisition. When multiple cues jointly predict an outcome, the most salient cue becomes associated with it and less salient cues are overshadowed. This asymmetry in cue saliency fosters selective attention, attenuating the impact of weaker cues during learning (cf. Kruschke & Blair, Reference Kruschke and Blair2000; Le Pelley et al., Reference Le Pelley, Haselgrove and Esber2012). Key evidence for this account comes from L2 studies showing learners’ difficulty in acquiring and applying linguistic features unique to the target language, especially when such features are absent from the L1 (Chen et al., Reference Chen, Shu, Liu, Zhao and Li2007; Dussias et al., Reference Dussias, Kroff, Tamargo and Gerfen2013; Ellis & Sagarra, Reference Ellis and Sagarra2010; Hou, Reference Hou2021; Luk & Shirai, Reference Luk and Shirai2009; Paul & Grüter, Reference Paul and Grüter2016).
Building upon theories of cue competition and differential weighting, specifically concerning socio-pragmatic, semantic, and structural cues, we employ Korean subject honorification as an empirical testbed. Although grammatically salient, subject honorification has low reliability and availability in real-life use (see Section 1.3), making it a structurally weak cue that competes with pervasive semantic–pragmatic information, especially given the absence of analogous grammatical marking in English. We examine how English-speaking learners navigate this cue asymmetry across offline and online tasks – production, perception, and processing – to assess how L2 learners integrate structurally encoded yet functionally variable cues in acquiring Korean.
1.2. Subject honorification in Korean
Among the two main dimensions of Korean honorification – referent honorifics (reflecting speaker–referent relationships) and addressee honorifics (reflecting speaker–addressee relationships) – both subject referent and addressee honorification are expressed through verbal suffixes (Sohn, Reference Sohn1999). The speaker’s respectful stance toward the subject referent is linguistically encoded by the subject honorific suffix -(u)si on a verb stem. Speakers’ deference, politeness, or condescension toward the addressee is further expressed through six levels of addressee honorification, indicated by sentence-final verbal endings. Because sentence-final predicate endings are obligatory in Korean grammar, the speaker’s stance toward the addressee is always encoded – whether honorific or non-honorific – through one of these six levels. For example, in (1), honorific agreement is observed between the subject (halmeni “grandmother”) and its predicate with -(u)si. In addition, because the sentence is addressed directly to the grandmother, the speaker’s politeness toward her is conveyed through -eyo/ayo, a polite sentence ender.
Previous studies on the processing of subject honorification have shown that native Korean speakers are sensitive to honorific feature agreement. When the subject honorific suffix is used with a non-honorifiable subject (e.g., kkoma “kid”) as in (2), this mismatch elicits processing difficulty, reflected in longer reading times (Kwon & Sturt, Reference Kwon and Sturt2016, Reference Kwon and Sturt2019) or a P600 response (Kwon & Sturt, Reference Kwon and Sturt2024) – a brain response also observed for number and person agreement violations in English and other languages (Hinojosa et al., Reference Hinojosa, Martín-Loeches, Casado, Munoz and Rubia2003; Osterhout & Mobley, Reference Osterhout and Mobley1995).
However, agreement in Korean subject honorification is context-dependent, unlike English number/person agreement. Speakers may omit the subject honorific suffix with honorifiable subjects without loss of acceptability as in (3). This flexibility arises when formality/hierarchy between interlocutors is relaxed, or when speaker−addressee relations are sufficiently close (Choo & Kwak, Reference Choo and Kwak2008; Sohn, Reference Sohn1999).
Honorifics can be omitted even in formal contexts (e.g., newspaper articles): honorifiable subjects take the subject honorific suffix only 41% in written texts (Kim et al., Reference Kim, Sells and Yang2005) and 37% in spoken texts (Song et al., Reference Song, Choe and Oh2019). This high optionality indicates that an honorifiable subject is an unreliable cue for the suffix. Cue availability is also low – 0.23 per ejels Footnote 2 (33,794 of 15,891,623 ejels; Song et al., Reference Song, Choe and Oh2019). Even when employed, 64% of instances target a subject referent who is also the addressee, and only 35% apply to a sentential subject distinct from the addressee (Song et al., Reference Song, Choe and Oh2019). Accordingly, low cue strength and frequent co-occurrence of subject and addressee honorification complicate acquisition of the suffix’s specific function and appropriate use, particularly for learners whose L1 lacks a comparable honorific system, such as English.
Furthermore, subject honorification can sometimes be applied to an inanimate, thus non-honorifiable, subject that is contextually associated with an honorific entity (inanimate honorification: Brown, Reference Brown2022; Choo & Kwak, Reference Choo and Kwak2008), as in (4).
Although inanimate honorification may be considered a grammatical error (Kim-Renaud, Reference Kim-Renaud and McAuley2001), it is increasingly attested in real-life usage, particularly across customer-service sectors, where it is used to signal respect to an addressee who holds interactional power (Eom, Reference Eom2019; Kwon & Lee, Reference Kwon and Lee2024). This practice appears to affect native speakers’ processing of the subject honorific suffix (Kim & Shin, Reference Kim and Shin2021), suggesting a functional extension toward a stand-alone politeness marker in certain contexts.
Subject honorification is socially grounded and semantically salient but exhibits low cue strength because of limited availability and complex structural requirements. It is typologically rare and absent from well-studied languages such as English. This constellation makes it an ideal testbed for investigating L2 learning mechanisms. However, relatively little research has examined its acquisition in either L1 or L2 Korean. For L1 acquisition, Jang (Reference Jang2018) conducted a longitudinal case study of a monolingual child (16–48 months), tracking the emergence of pre-final sentence enders, including the subject honorific suffix. The child produced this suffix as early as 17 months in fully conjugated predicates (e.g., cwu-si-eyo give-sh-dcl.ah “(please) give (it to me)”) and showed more productive use by around 25 months. Despite this early emergence, subject honorification developed later than addressee honorification. Although based on a single participant, this relatively delayed trajectory may reflect the low cue strength of subject honorification.
Concerning L2 acquisition, Mueller and Jiang (Reference Mueller and Jiang2013) tested whether advanced L1-English L2-Korean learners are sensitive to subject honorification in a self-paced reading task manipulating the honorific features of the subject and predicate. Although participants demonstrated explicit knowledge of the subject honorific suffix on a written assessment, they showed no online sensitivity to mismatches between the subject and predicate, supporting the view that L2 processing is more limited online than offline. In contrast, Kim et al. (Reference Kim, Kim and Kim2025) showed that learners’ online sensitivity improves when they are explicitly instructed to anticipate upcoming referents using honorific features as a cue, suggesting that prediction-oriented, form-focused goals can enhance real-time use of the relevant grammatical cues.
While prior work indicates that L2 learners of Korean can develop some sensitivity to subject honorification, it remains unclear whether this sensitivity aligns with its grammatical function. Given its low cue strength and frequent co-occurrence with addressee honorification, the distinct function of subject honorification may not be fully acquired; learners may instead construe and deploy it as a general politeness device directed at the addressee. In this study, we test this possibility using both offline and online tasks.
1.3. Current study
Theoretical research on Korean has long highlighted the social and grammatical significance of subject honorification. Within L2 acquisition contexts, the subject honorific suffix has often been overly emphasized in formal instructional settings, leading to a disproportionate focus on this suffix in classroom materials (Brown, Reference Brown2010; Jung et al., Reference Jung, Shin and Lee2025). However, despite this heavy emphasis, learners – even at advanced levels – continue to struggle with its acquisition (Mueller & Jiang, Reference Mueller and Jiang2013), although explicit task goals may facilitate more effective online processing (Kim et al., Reference Kim, Kim and Kim2025). The low cue strength of subject honorification, combined with its frequent co-occurrence with addressee honorification in native speakers’ usage (Song et al., Reference Song, Choe and Oh2019) could further complicate learners’ ability to distinguish and apply each system independently.
With these considerations in mind, we pose the following research question: How do L1-English learners of Korean produce, perceive, and process subject honorification? To address this question comprehensively, we adopt a multi-method design comprising three complementary studies. Study 1 analyzes open-access corpus data to examine learners’ usage patterns, revealing spontaneous usage strategies of subject honorification in production. Study 2 employs a controlled politeness-rating task with recruited participants to assess learners’ explicit socio-pragmatic understanding of honorific use in an offline manner. Study 3 adopts a self-paced reading paradigm, administered to the same cohort using modified stimuli from Study 2, to investigate online processing of subject–predicate honorific agreement. By encompassing production (Study 1), metalinguistic evaluations (Study 2), and real-time processing (Study 3), our approach offers an integrated picture of L1-English L2-Korean learners’ knowledge about Korean subject honorification.
2. Study 1: Corpus analysis
We explored L2-Korean learners’ use of honorifics using the Korean Learners’ CorpusFootnote 3, supplemented with learners’ L1, country of origin and proficiency (Levels 1–6, following the scale set by the Test of Proficiency in Korean). We analyzed 135,117 tagged ejels from 1,094 L1-American English learners across all levels (Level 1: 239; Level 2: 259; Level 3: 206; Level 4: 168; Level 5: 134; Level 6: 85; unclassified: 3), encompassing written and spoken production data. We extracted all sentences containing subject honorification, yielding a subset from 200 learners (Level 1: 56; Level 2: 64; Level 3: 30; Level 4: 26; Level 5: 13; Level 6: 9; unclassified: 2), comprising 395 written and 52 spoken tokens. Sentences were manually evaluated for accuracy; incorrect usages were identified solely on the use of the subject honorific suffix, not spelling. The remaining sentences were classified by the target of subject honorification. Given the frequent co-occurrence of subject and addressee honorification in native speaker data, we hypothesized that learners may not have fully acquired their independent functions; accordingly, we coded sentences as featuring both subject and addressee honorification or subject honorification only (Table 1). Figure 1 presents correct and incorrect usage across proficiency levels (see Supplementary Table S1 for details).
Results (Study 1): Corpus analysis of subject honorification use
Note: Accuracy was evaluated based on the use of subject honorific suffix, not on spellings. See Supplementary Table S1 for detailed counts.
Correct and incorrect use of subject honorification by L1-English L2-Korean learners across proficiency levels.
We found that learners at the lowest proficiency level produced the most instances of subject honorification: 191 utterances with the subject honorific suffix out of 21,846 ejels (0.87% per ejel), with only 10 clear errors. Notably, its usage decreased with proficiency levels: Level 2, 104 utterances out of 29,933 ejels (0.35% per ejel); Level 3, 60 utterances out of 26,696 ejels (0.22% per ejel); Level 4, 45 utterances out of 20,416 ejels (0.22% per ejel); Level 5, 27 utterances out of 23,096 ejels (0.12% per ejel); Level 6, 23 utterances out of 12,750 ejels (0.18% per ejel).
A chi-square test of independence revealed a significant association between proficiency level and subject honorification usage, χ 2(5, N = 134,737) = 248.84, p < 0.001, indicating variation across proficiency levels. The lowest proficiency group used subject honorification most, with a sharp decline among more advanced proficiency groups. Although this might suggest early mastery, the frequent use at low proficiency more plausibly reflects overgeneralization rather than a nuanced command of the Korean honorific system.
We also found that subject and addressee honorification frequently co-occurred. To test whether co-occurrence varied by proficiency, we performed a chi-square test of independence on collapsed proficiency groups (Levels 1–2: low; Levels 3–4: intermediate; Levels 5–6: advanced). This reclassification was implemented to avoid violations of the minimum expected cell count requirement (McHugh, Reference McHugh2013) caused by the uneven distribution in the original six proficiency levels, thereby satisfying statistical assumptions while clarifying proficiency-related patterns. The results revealed a significant effect of proficiency levels: χ 2(2, N = 425) = 101.56, p < .001, indicating that co-occurrence was highest in the low group, dropped sharply in the intermediate group, and decreased further in the advanced group. Bonferroni-corrected pairwise comparisons confirmed differences between low versus intermediate (χ 2(1) = 47.04, p < .001), low versus advanced (χ 2(1) = 97.51, p < .001) and intermediate versus advanced (χ 2(1) = 9.33, p < .007). Overall, the corpus analyses showed that the lowest-proficiency learners produced the most subject honorification – and almost always alongside addressee honorification – whereas both usage and co-occurrence declined with proficiency.
These findings suggest that learners may be replicating the co-occurrence common in native speaker usage. Although subject and addressee honorification serve independent functions, native speaker data indicate frequent interactions between the two, particularly in conversational settings where socio–pragmatic pressures shape honorific choices. Subject honorification is optional (Kim et al., Reference Kim, Sells and Yang2005; Song et al., Reference Song, Choe and Oh2019), but its use is often motivated by politeness strategies and social face management, especially when the subject referent coincides with the addressee (Kwon & Lee, Reference Kwon and Lee2024). Corpus evidence corroborates this: 64% of subject honorific tokens target subjects who are also the addressee (Song et al., Reference Song, Choe and Oh2019). For learners, this overlap poses a challenge because both systems are realized in verb conjugation, making their independent functions harder to differentiate and likely contributing to the high co-occurrence observed at low proficiency.
Concerning proficiency, our findings suggest that learners may develop greater awareness of the complexities of the honorific system as proficiency increases. Alternatively, reduced use may reflect hesitation or uncertainty in applying these forms as distinctions between subject and addressee honorification become clearer, or a perception that subject honorification is redundant when addressee forms are used – particularly before their independent functions are fully acquired.
3. Study 2: Politeness rating
Building upon the findings of Study 1 that L2 learners struggle with interconnected honorific systems in Korean, we directly assessed their understanding of subject honorification using a politeness rating task. This method probes explicit interpretations of honorific usage, capturing metalinguistic awareness and socio-pragmatic evaluation that production data may not reveal. To this end, we designed four target conditions by crossing subject type (honorifiable versus non-honorifiable) with predicate type (honorific versus non-honorific) and added two controls with inanimate subjects (one per predicate type) to induce semantic anomaly. Crucially, no sentence contained addressee honorification. Given the optionality of the subject honorific suffix, an honorific subject is compatible with both predicate types, whereas a non-honorifiable subject is compatible only with a non-honorific predicate. Native Korean speakers show longer reading times (Kwon & Sturt, Reference Kwon and Sturt2016, Reference Kwon and Sturt2019) and P600 responses (Kwon & Sturt, Reference Kwon and Sturt2024) to such honorific violations, indicating that these are processed as structural anomalies. Inanimate subjects are incompatible with the predicates used in Study 2 irrespective of subject honorification. Table 2 summarizes the conditions and their acceptability.
Types of subject–predicate combinations and their acceptability
Based on these conditions, we derive the following predictions. If learners have acquired subject honorification, sentences with a non-honorifiable subject and an honorific predicate should receive significantly lower ratings than their non-honorific counterparts (i.e., a grammaticality effect). Ratings may also be lower for an honorific subject with a non-honorific predicate if learners are sensitive to textbook-emphasized politeness norms (Brown, Reference Brown2010; Jung et al., Reference Jung, Shin and Lee2025), although this effect should be weaker than the main violation. Finally, sentences with inanimate subjects should receive low ratings regardless of predicate type because of semantic anomaly; this effect should obtain irrespective of learners’ successful acquisition of subject honorification.
3.1. Methods
3.1.1. Participants
We recruited 42 English-speaking learners of Korean (ELK; Mage = 24, SDage = 5), who were all non-heritage speakers currently residing in the United States and enrolled in third-year or higher university-level Korean courses. These inclusion criteria ensured a relatively homogeneous learner profile while preserving the proficiency variability essential to our research aims. At the time of testing, participants reported spending varying amounts of time abroad in South Korea (Mmonths = 17, SD = 31; min months = 0, max months = 130) and studying Korean at universities (Myear = 4, SD = 3; min years = 0, max years = 10). Twenty native speakers of Korean (NSK; Mage = 26, SD = 5) were also recruited as a control group. Informed consent was obtained from all participants, and the study protocol was approved by the Institutional Review Board to ensure compliance with ethical research standards.
Proficiency in Korean was assessed using the Korean C-test (Lee-Ellis, Reference Lee-Ellis2009), which evaluates comprehension of Korean sentences of varying lengths and complexities. The test consists of five passages with syllable-unit blanks; for testing efficiency, we selected the first four passages, as recommended by the original study. Each blank corresponded to one point, and the maximum score possible was 188. The mean score of ELK was 101, and the standard deviation was 41.
3.1.2. Stimuli
We created 36 sets of experimental sentences as in (5). To keep the stimuli simple and common, we selected nouns (e.g., kinship terms, proper names, occupations), adjectives, and verbs from the most frequent words in the 6,230,590 untagged tokens of the Korean Learners’ Corpus3. The mean word frequency was 2,931 (maximum: 24,181; minimum: 0); the two zero-frequency items were proper nouns. The full list of test sentences is available in the OSF repository.
Prior to the experiment, we normed the test sentences by having 18 native Korean speakers (who did not participate in the main experiment; Mage = 26, SDage = 5), assessing the naturalness of the events described in these sentences using a 6-point Likert scale (0 = unnatural; 5 = natural). The target items for the norming study were created for both Study 2 (politeness rating) and Study 3 (self-paced reading), with the subject honorific suffix removed from the NHsubj and INAsubj condition sentences in both studies (see Supplementary Table S2 for the results of the norming study). Bonferroni-corrected pairwise comparisons of the norming results revealed significant differences between the Hsubj and INAsubj conditions and between the NHsubj and INAsubj conditions (all ps < 0.001). However, the Hsubj and NHsubj conditions did not differ significantly. These results indicate that the events described in the INAsubj condition are perceived as less natural compared to those in the other two conditions. The results also suggest that the target sentences in the Hsubj and NHsubj conditions did not differ in the naturalness of the events described. Therefore, the norming results confirmed that the target sentences for Studies 2 and 3 were created as intended.
Based on the norming results, the sentences were divided into six lists (reflecting a Latin square design) and randomly assigned to participants, with the presentation order within each list also randomized. To obscure the purpose of the experiment, each list also included 72 filler sentences of varying complexity and acceptability.
3.1.3. Procedure
We administered the task using Qualtrics, a web-based survey platform. Participants received an individual link and were instructed to rate the politeness of each sentence on a 6-point Likert scale (0: very impolite; 5: very polite) as quickly as possible while maintaining the faithfulness and confidence of their responses. Once a participant selected a rating and proceeded to the next sentence, they could not revisit or alter previous responses. To ensure data quality, participation via mobile devices was not permitted. The task duration was approximately 20 minutes.
3.1.4. Analysis
Data from one participant were removed due to incompleteness. Remaining data were analyzed using ordinal regression models with the cumulative link mixed model function from the ordinal package (Christensen, Reference Christensen2023) in R (R Core Team, 2024). The regression model included three fixed-effect factors: Group (NSK, ELK), Subject (Hsubj, NHsubj, INAsubj) and Predicate (Hpred, NHpred), as well as with their interactions. The Group factor was treatment-coded, with NSK serving as the reference level, while both Subject and Predicate were sum-coded. Upon observing a significant interaction involving Group, we conducted separate analyses for the NSK and ELK groups. In the NSK model, only Subject and Predicate were included. In contrast, the ELK model additionally included Proficiency, the scores measured through the Korean C-test, as a continuous variable to examine potential developmental effects among L2 learners. Random intercepts for participants and items were included to account for variability, as permitted by the design and model convergence (cf. Barr et al., Reference Barr, Levy, Scheepers and Tily2013). Given the ordinal nature of the dependent variable, a logit link function was applied to model the cumulative probability of higher ratings. Model fitting and validation included checks for convergence, proportional odds assumptions and residual diagnostics. Post-hoc pairwise comparisons were conducted using emmeans (Lenth, Reference Lenth2025). The p-values were derived from Wald z-tests, as implemented in the ordinal package (Christensen, Reference Christensen2023). When significant interactions with Group were observed, we conducted separate follow-up analyses for the NSK and ELK groups to clarify the nature of these interactions.
3.2. Results
Figure 2 presents politeness ratings by group and condition. The two groups showed a similar pattern of ratings for the Hsubj condition, with both identifying the Hsubj_Hpred condition as the most polite. However, their evaluations diverged for the NHsubj and INAsubj conditions.
Mean politeness ratings for NSK (native speakers of Korean) and ELK (English-speaking learners of Korean). Error bars indicate 95% confidence intervals.
The global analysis (Table 3a) showed main effects of Subject and Predicate; these main effects were moderated by two-way interactions between Group and Subject, Group and Predicate and Subject and Predicate, alongside a three-way interaction. Bonferroni-corrected pairwise comparisons (Appendix S1) revealed that, within the Hpred conditions, the two groups differed significantly in the NHsubj (z = −3.67, p < .002) and INAsubj (z = −5.40, p < .001) conditions, while the group difference in the Hsubj condition was marginal (z = 2.62, p = .053). The significant differences in the NHsubj and INAsubj conditions reflected higher ratings by the ELK group compared to the NSK group, whereas the marginal difference in the Hsubj condition was driven by higher ratings from the NSK group. Within the NHpred conditions, no significant group differences emerged in the Hsubj (z = 0.78, p = 1.00) and INAsubj (z = −1.98, p = .28) conditions. However, in the NHsubj condition, the NSK group assigned significantly higher ratings than the ELK group (z = 2.91, p = .021).
Model outputs (Study 2): Cumulative link mixed-effects model results for politeness rating. (a) Global model including all groups with interaction terms; (b) Model for NSK (native speakers of Korean); (c) Model for ELK (English-speaking learners of Korean)
Note: Fixed effects included Subject (honorifiable, non-honorifiable, Inanimate), Predicate (honorific, Non-honorific), Group (NSK, ELK), and their interactions. In the ELK model, Proficiency was also included as a continuous variable. Thresholds represent intercepts between adjacent response categories on the 6-point rating scale. Indicators in the Slope column specify whether the corresponding factor was included as a random slope for participants (p) or items (i).
To zoom into each group’s rating performance, we examined the data separately for the NSK and ELK groups. The NSK model (Table 3b) revealed main effects of Subject, with higher ratings for the Hsubj condition than the NHsubj condition, which in turn elicited higher ratings than the INAsubj condition. There was also a main effect of Predicate, with higher ratings for the Hpred condition than the NHpred condition. These main effects were qualified by interactions between them. Bonferroni-corrected pairwise comparisons (Appendix S1) revealed that the Hpred condition was rated significantly more polite than the NHpred condition when the subject was honorifiable (z = 8.13, p < .001). When the subject was non-honorifiable or inanimate, there was no effect of predicate types (all ps < .6).
The ELK model (Table 3c) revealed a significant main effect of Predicate and a Predicate × Proficiency interaction. The main effect of Predicate was driven by lower ratings for sentences in the NHpred condition than those in the Hpred condition, irrespective of subject type (Appendix S1). To unpack the interaction, we examined how proficiency influenced ratings within each predicate condition using the emtrends() function (Lenth, Reference Lenth2025). The results showed that, while Proficiency did not significantly affect ratings for sentences in the Hpred condition, higher proficiency yielded lower politeness ratings in the NHpred condition. These findings indicate that, as proficiency scores increased, learner participants assigned lower politeness ratings to sentences with non-honorific predicates, while ratings for sentences with honorific predicates remained stable across proficiency scores. There was also a marginal Subject × Predicate × Proficiency interaction. We conducted an additional emtrends() analysis to assess how the proficiency-related slope differences between honorific and non-honorific predicate conditions varied by subject type. The results showed that, with increasing proficiency, the politeness-rating gap between Hpred and NHpred conditions widened for both Hsubj and NHsubj conditions, reflecting heightened sensitivity to the subject honorific suffix when the subject was animate. By contrast, proficiency did not significantly affect this distinction in the INAsubj condition.
3.3. Discussion
The NSK and ELK groups exhibited notable differences in their recognition of politeness across the subject and predicate types. For the NSK group, sentences in the Hpred condition were rated as more polite than those in the NHpred condition only when the subject was honorifiable, demonstrating no significant effect of predicate type for non-honorifiable or inanimate subjects. These findings suggest that native Korean speakers regard the subject honorific suffix as contributing to politeness only under its grammatically licensed use. In contrast, the ELK group rated sentences in the Hpred condition as more polite than those in the NHpred condition, regardless of the subject’s honorifiability, and this pattern persisted when taking proficiency into account. As proficiency increased, learner participants assigned progressively lower politeness ratings to sentences in the NHpred condition – widening the rating gap between Hpred and NHpred ratings for both honorifiable and non-honorifiable subjects – while their evaluations of sentences with inanimate subjects remained unchanged by proficiency scores.
L2 learners may have approached the task assuming that all sentences were grammatically acceptable, perhaps influenced by task framing or limited awareness of honorific constraints, and may have treated the subject honorific suffix as a reliable marker of politeness, a default heuristic in their evaluations. Under this assumption, they may have consistently rated Hpred sentences as more polite, even when this suffix occurred with semantically or grammatically inappropriate subjects. Yet the consistency of their ratings across subject types – including inanimate cases – point to more than a transient heuristic: learners seem to have reinterpreted this suffix not as morphosyntactic agreement marker contingent on subject honorifiability but as a broad socio-pragmatic signal of politeness. In this regard, their interpretation suggests a pragmatic reanalysis of this suffix, revealing a gap between emerging socio-pragmatic competence and the precise morphosyntactic constraints of honorification in Korean.
4. Study 3: Self-paced reading
The politeness rating task in Study 2 revealed that L1-English L2-Korean learners’ understanding of subject honorification is limited: despite extensive textbook coverage, learners tended to interpret the subject honorific suffix as a general politeness marker, attending less to its specific grammatical function tied to the subject’s honorifiability. Study 3 builds upon these findings, by testing the same learner group with revised stimuli in a self-paced reading task to examine real-time processing of subject honorification. For the native speaker group, we recruited 42 additional participants naïve to the design. The self-paced reading paradigm – widely used to assess incremental processing difficulty in dependency resolution (Just et al., Reference Just, Carpenter and Woolley1982; Kaiser, Reference Kaiser, Podesva and Sharma2014) and L2 reading processes (Keating & Jegerski, Reference Keating and Jegerski2015; Marsden et al., Reference Marsden, Thompson and Plonsky2018) – yields effects comparable to higher-resolution methods such as eye-tracking (e.g., Kwon & Sturt, Reference Kwon and Sturt2016). Combined with web-based data collection,Footnote 4 it enables broad recruitment without geographic constraints. By comparing learners’ offline politeness ratings with their online sensitivity to honorific violations, Study 3 tests whether overgeneralized interpretations surface in real-time processing, offering a more comprehensive view of developing grammatical competence.
To more directly assess learners’ sensitivity to the grammatical function of subject honorification during online processing, we introduced a minor modification to the target stimuli in Study 3. Specifically, we embedded a cataphoric dependency by placing the subject after the predicate as in (6), to create a context in which comprehenders must anticipate the honorific features of the upcoming subject based solely on the presence or absence of the honorific suffix on the predicate.
Under these experimental conditions, we can formulate three predictions regarding learners’ performance on this task. First, in the Inanimate Subject condition, both honorific-predicate sentences (6a) and non-honorific-predicate sentences (6b) should elicit increased processing difficulty at the inanimate subject position (e.g., picture), as predicates such as tuleka- (“to enter”) semantically require an animate subject, prompting comprehenders to anticipate animacy regardless of honorification knowledge. Second, in the Non-honorifiable Subject condition, learners who have successfully acquired the subject honorific suffix should exhibit greater processing difficulty at the subject (e.g., “Mia”) when it is preceded by an honorific predicate (6a) than by a non-honorific predicate (6b), reflecting a violation of the grammatical requirement that the subject honorific suffix co-occurs only with honorifiable subjects. Third, in the Honorifiable Subject condition, neither honorific nor non-honorific predicates (6a or 6b) should incur processing cost at the subject position (e.g., “teacher”), as the use of subject honorific suffix is optional with honorifiable referents.
4.1. Methods
4.1.1. Participants and stimuli
For the ELK group, the same participants completed this task after finishing Study 2. For the NSK group, a new set of 40 participants (M age = 26, SD = 5) was recruited to complete the task. Thirty-six sets of target sentences were divided into six lists following a Latin square design, with sentences presented in a random order within each list. Seventy-two filler sentences of varying structures and complexities were included to obscure the purpose of the experiment.
4.1.2. Procedure
The sentences were presented region by region using a non-cumulative moving-window paradigm (Just et al., Reference Just, Carpenter and Woolley1982) through PCIbex (Zehr & Schwarz, Reference Zehr and Schwarz2018). After each sentence, a simple comprehension question appeared to ensure participants’ attention to the task (e.g., questions about the sentence’s subject, the action performed, or the relationships between entities). Participants responded by selecting either “yes” or “no” with feedback provided as “wrong choice” for incorrect responses. Prior to the experiment, participants received written instructions and completed three practice items to become familiar with the procedure. The task took approximately 25 minutes.
4.1.3. Analysis
Data from four participants were excluded due to a software failure. The remaining data were initially trimmed by excluding reading time data points below 150 ms or above 4000 ms. The remaining data were log-transformed and further trimmed by removing points exceeding three standard deviations from the condition mean (data loss: ELK: 2.58%; NSK: 1.52%). As lexical items at R4 differed across conditions, log-transformed reading times at R4 were residualized to control for variability in word length and individual reading speed (Baayen & Milin, Reference Baayen and Milin2010). R4 serves as the critical region, with R5 and R6 included to account for spillover effects from the task-specific button-press strategy (Koornneef & Van Berkum, Reference Koornneef and Van Berkum2006). All remaining analytical procedures matched Study 2; preprocessed data were fitted to linear mixed-effects models using the lme4 package (Bates et al., Reference Bates, Mächler, Bolker and Walker2015).
4.2. Results
Comprehension accuracy was high for both the NSK (M = 97, SD = 0.03, Max = 100, Min = 89) and ELK (M = 89, SD = 0.1, Max = 100, Min = 55) groups, indicating that participants were attentive to the task. For the reading-time analysis, only trials with correct answers were included. Reading-time results for the target regions are presented in Figures 3–5.
Mean residual reading times at R4 for NSK (native speakers of Korean) and ELK (English-speaking learners of Korean). Error bars represent standard errors.
Mean raw reading times (trimmed) at R5 for NSK (native speakers of Korean) and ELK (English-speaking learners of Korean). Error bars represent standard errors.
Mean raw reading times (trimmed) at R6 for NSK (native speakers of Korean) and ELK (English-speaking learners of Korean). Error bars represent standard errors.
4.2.1. Global analysis
Statistical analyses were conducted on reading times at the critical region (R4) and two spill-over regions (R5, R6). Table 4 presents model outputs.
Model outputs (Study 3): Linear mixed-effects models for reading times (R4, R5 and R6) involving both NSK (native speakers of Korean) and ELK (English-speaking learners of Korean)
Note: The models were fitted separately for each region. Fixed effects included Subject (honorifiable, non-honorifiable, Inanimate), Predicate (honorific versus non-honorific), Group (NSK versus ELK), and their interactions. Random intercepts for participants and items were included. Indicators in the Slope column specify whether the corresponding factor was included as a random slope for participants (p) or items (i).
At R4, we found a main effect of Predicate, accompanied by an interaction effect between Group and Subject and a marginal interaction effect between Subject and Predicate. Bonferroni-corrected pairwise comparisons (Appendix S2) showed no significant group differences for any of the conditions (all ps < .7). At R5, we found main effects of Group and Subject, alongside a three-way interaction effect between Group, Subject, and Predicate. Bonferroni-corrected pairwise comparisons (Appendix S2) revealed that the NSK group was significantly faster than the ELK group across all six conditions (all ps < .001). At R6, we found main effects of Group and Subject, accompanied by a three-way interaction effect between Group, Subject, and Predicate. Bonferroni-corrected pairwise comparisons (Appendix S2) revealed that the NSK group were significantly faster than the ELK group across all six conditions (all ps < .001).
4.2.2. By-group analysis
To further examine each group’s reading-time performance, we analyzed the data separately for the NSK and ELK groups, as presented in Table 5.
Model outputs (Study 3): Linear mixed-effects models for reading times (R4, R5 and R6) for NSK (native speakers of Korean) and ELK (English-speaking learners of Korean)
Note: The models were fitted separately for each region. Fixed effects included Subject (honorifiable, non-honorifiable, inanimate), Predicate (honorific versus non-honorific) and their interactions. Random intercepts for participants and items were included. Indicators in the Slope column denote whether the corresponding factor was included as a random slope for participants (p) or items (i).
The NSK model revealed a main effect of Predicate at R4, with longer reading times for the Hpred condition (M = 0.023, SE = 0.017) compared to the NHpred condition (M = −0.023, SE = 0.017). This effect was further moderated by an interaction effect with Subject. Bonferroni-corrected pairwise comparisons (Appendix S2) revealed that, within the INAsubj condition, the Hpred condition took significantly longer to read than the NHpred condition (t = 3.07, p = .003). No other effects were significant. At R5, a main effect of Subject was found. However, Bonferroni-corrected pairwise comparisons (Appendix S2) revealed no significant difference between the subject types. At R6, there was a main effect of Subject. Bonferroni-corrected pairwise comparisons (Appendix S2) revealed that reading times in the INAsubj condition were significantly longer than in both the Hsubj (t = −5.59, p < .001) and NHsubj conditions (t = 4.42, p < .001), which did not differ from each other. Moreover, a significant Subject × Predicate interaction emerged: for non-honorifiable subjects, sentences in the Hpred condition elicited longer reading times than those with NHpred condition (t = 2.23, p = .026), whereas predicate type had no effect on reading times for either honorifiable or inanimate subjects (all ps < .2). Together, the results indicate that (i) the Hsubj condition did not elicit longer reading times, regardless of predicate type, (ii) the NHsubj condition elicited longer reading times only when paired with an honorific verb (a grammaticality effect) and (iii) the INAsubj condition elicited longer reading times regardless of predicate type (a semantic anomaly effect).
Concerning the ELK model, we found a main effect of Proficiency at R4, with higher-proficiency learners reading R4 more quickly. Additionally, a Subject × Predicate interaction reached significance. Bonferroni-corrected pairwise comparisons (Appendix S2) revealed no significant Hpred–NHpred differences for either Hsubj or NHsubj conditions. In the INAsubj condition; however, the Hpred–NHpred contrast approached marginal significance (t = 1.94, p = .054), with faster reading times in the NHpred condition than the Hpred condition.
At R5, we found a main effect of Proficiency, with higher-proficiency learners reading faster. A marginal Subject × Predicate interaction prompted Bonferroni-corrected pairwise comparisons (Appendix S2): for the Hsubj condition, reading times were significantly longer in the NHpred condition than the Hpred condition (t = −2.28, p = .023), whereas no predicate-type differences emerged for the INAsubj or NHsubj conditions (all ps < .4). Additionally, a significant Subject × Proficiency interaction emerged because the rate of improvement was attenuated for inanimate subjects relative to the other conditions, even though reading times decreased with proficiency across all subject types (all ps < .001).
At R6, a main effect of Subject was observed: sentences with inanimate subjects elicited longer reading times than those with either honorifiable (t = −4.41, p < .001) or non-honorifiable subjects (t = 5.05, p < .001), and there was no difference between the Hsubj and NHsubj conditions (p = 1.00). Predicate type did not produce a significant main effect, while higher proficiency was associated with faster reading times overall. Although the Subject × Predicate interaction was insignificant, we conducted post-hoc comparisons (Appendix S2) to probe predicate effects within each subject type. For the Hsubj condition, sentences in the NHpred condition yielded longer reading times than those in the Hpred condition (t = −2.12, p = .035), whereas predicate type had no effect for the INAsubj or NHsubj conditions (all ps < .5). No other contrasts reached significance. Together, these findings at R6 indicate (i) a semantic anomaly effect in the INAsubj condition, (ii) heightened sensitivity to the subject honorific suffix in the Hsubj condition and (iii) the absence of a grammaticality effect in the NHsubj condition.
4.3. Discussion
The NLK group demonstrated both a semantic anomaly effect (i.e., longer reading times for the INAsubj condition) and a grammaticality effect (i.e., longer reading times for the NHsubj condition) when the subject honorific suffix was present. For the Hsubj condition, the presence or absence of this suffix did not induce processing difficulty, consistent with the optionality of subject honorification in Korean.
The ELK group likewise showed a semantic anomaly effect (i.e., longer reading times for the INAsubj condition) but, unlike the NSK group, no grammaticality effect: the subject honorific suffix did not lengthen reading times for the NHsubj condition. Even so, they exhibited some awareness of honorific expressions: reading times increased at R4 when inanimate NPs were paired with honorific verbs, and at R5–R6 when honorific subjects lacked honorific verbs. These patterns suggest partial sensitivity to honorific predicates but incomplete acquisition of subject honorification.
5. General discussion
Korean subject honorification illuminates the intricate interplay of structural, semantic, and socio-pragmatic factors, making it an ideal case for exploring cue competition and differential weighting in L2 acquisition. Although superficially akin to number/person agreement in languages such as English, it differs fundamentally: the subject honorific suffix is not obligatorily triggered by an honorifiable subject but rather hinges upon socio-pragmatic cues, relative hierarchy and the speaker’s stance toward the referent. Corpus analyses confirm its optionality – only about 40% of honorifiable subjects co-occur with honorific predicates (Kim et al., Reference Kim, Sells and Yang2005; Song et al., Reference Song, Choe and Oh2019) – marking it as a structurally weak cue. Nevertheless, the incorrect use of the subject honorific suffix with a non-honorifiable subject substantially increases processing costs, yielding longer reading times (Kwon & Sturt, Reference Kwon and Sturt2016, Reference Kwon and Sturt2019) and a P600 response (Kwon & Sturt, Reference Kwon and Sturt2024), paralleling effects in number/person agreement violations (Hinojosa et al., Reference Hinojosa, Martín-Loeches, Casado, Munoz and Rubia2003; Osterhout & Mobley, Reference Osterhout and Mobley1995). Acquisition is further complicated by the rise of inanimate honorification in service sectors, adding unpredictability. Pedagogical materials also tend to over-present subject honorification without adequate context, obscuring its context-dependent nuances (Brown, Reference Brown2010; Jung et al., Reference Jung, Shin and Lee2025). Collectively, these factors pose considerable challenges for L2-Korean learners.
5.1. Implications of findings
In this study, we investigated L1-English L2-Korean learners’ acquisition of subject honorification through three complementary studies: corpus analysis (Study 1), politeness rating (Study 2) and self-paced reading experiments (Study 3). Study 1 revealed that lower-proficiency learners frequently used subject honorification alongside addressee honorifics, with co-occurrence decreasing as proficiency increased, aligning with native speakers only at advanced proficiency levels. This suggests that isolated use of subject honorification may not reliably indicate successful acquisition. These results support the semantics-before-syntax account (e.g., Gass, Reference Gass1986; Grüter et al., Reference Grüter, Lau and Ling2020), implying that prominent semantic or pragmatic cues may suppress less salient morphosyntactic distinctions during L2 acquisition.
Results from the politeness rating experiment (Study 2) further support this interpretation (although the target populations differed). Learners consistently perceived sentences with the subject honorific suffix as significantly more polite, regardless of the honorifiability of the subject, unlike native speakers who associated increased politeness exclusively with honorifiable subjects. Proficiency appeared to influence learners’ awareness to a limited extent: higher-proficiency learners assigned lower politeness ratings to conditions lacking the subject honorific suffix than lower-proficiency learners, irrespective of subject type. These findings suggest that learners may predominantly interpret the suffix as a general politeness marker rather than a specific marker for morphosyntactic agreement. This resonates with L2 learners’ primary reliance on semantic-pragmatic cues during L2 activities.
The self-paced reading experiment (Study 3) extends these findings. We found that both learner and native speaker groups exhibited sensitivity to semantic anomalies involving inanimate subjects (e.g., R6). In addition, learners, similar to native speakers, showed heightened sensitivity to honorifics – especially with inanimate subjects and honorifiable subjects. However, unlike native speakers, learners showed no sensitivity when processing morphosyntactic violations (i.e., non-honorifiable subjects accompanied by honorific predicates). This suggests that learners preferentially employ shallow, semantic-based processing over detailed grammatical analyses unless explicitly guided (Lim & Christianson, Reference Lim and Christianson2013; Omaki & Schulz, Reference Omaki and Schulz2011), further aligning with the semantics-before-syntax accounts.
Although Study 1, the corpus study based on data from 200 Korean learners, sampled a different participant population than Studies 2 and 3 – thus warranting caution when integrating their findings – the results were remarkably consistent: learners exhibited some awareness of honorific marking but struggled with the independent function of subject honorification. This interpretation aligns with corpus analysis results showing that learners predominantly used subject honorification alongside addressee honorification. The politeness rating results further support this view, revealing that learners perceived increased politeness with the subject honorific suffix, regardless of subject type. This is also consistent with the self-paced reading results, where learners exhibited processing difficulty only in response to semantic anomalies but not ungrammaticality. These consistent findings across distinct participant populations and task types strengthen our overall argument: English-speaking learners of Korean may perceive and apply the subject honorific suffix as a general politeness marker rather than as a specific indicator of subject honorification, underscoring L2 learners’ greater reliance on salient semantic and pragmatic cues over morphosyntactic information.
5.2. Broader implications to L2 development
Learners’ limited grasp of subject honorification likely arises from the relatively low cue validity of honorifiable subjects in signaling the subject honorific suffix. This weak cue strength, combined with exposure to the frequent co-occurrence between subject and addressee honorification and inanimate honorification, appears to hinder the effective acquisition of the independent function of subject honorification. This challenge persists alongside the heavy emphasis on the subject honorific suffix for subject honorification (and Korean honorific systems in general) in Korean language textbooks, which could offer an initial/default strategy for honorifics. While such emphasis may promote awareness, acquiring a complex honorific system remains difficult for learners whose L1 lacks a comparable system such as English, the L1 of our target population in this study.
In this context, learners’ perception and use of the subject honorific suffix as a general politeness marker support the overshadowing-and-blocking account (Ellis, Reference Ellis2006, Reference Ellis, Robinson and Ellis2008). Successful acquisition of subject honorification requires learners’ attention to diverse cues, including the absence of the subject honorific suffix with non-honorifiable subjects (i.e., morphosyntactic agreement cues), and its independence from addressee honorification, both of which are essential for accurate structural analysis. However, these cues may have been overshadowed by more salient semantic and socio–pragmatic factors involving the use of this suffix. Together with exposure to the subject honorific suffix with its extended usage, such as inanimate honorification in the customer service sector, learners’ primary reliance on semantic cues over structural cues (Gass, Reference Gass1986; Grüter et al., Reference Grüter, Lau and Ling2020) may have rendered them less efficient – and potentially oblivious – to the precise handling of sentences involving this knowledge. Nevertheless, learners demonstrate partial grammatical computations in specific contexts, as observed in their real-time processing patterns at R4 when inanimate subjects were associated with honorific verbs. Given multiple-stream models that posit both shallow and full parsing routes (Clahsen & Felser, Reference Clahsen and Felser2018; Christianson, Reference Christianson2016; Ferriera, 2003), this implies that learners could compute the honorific agreement relation when explicitly required (albeit inconsistent and inefficient).
Our findings also shed light on the nuanced role of proficiency in L2 acquisition. The modest impact of proficiency on learner performance found in this study contrasts with prior research highlighting its facilitative effects that ensure target-like processing (Hopp, Reference Hopp2017; Jackson, Reference Jackson2008; Rah & Adone, Reference Rah and Adone2010). This discrepancy hints at two possibilities. First, proficiency effects may differ depending on task types in L2 activities (Roberts, Reference Roberts2012). Second, akin to L1 speakers, L2 learners may employ good-enough processing, prioritizing early interpretation over correcting misinterpretations unless essential (Christianson, Reference Christianson2016; Ferreira, Reference Ferreira2003). It is also possible that grammatical functions with low cue strength, such as subject honorification, require more advanced proficiency for successful acquisition than our participants currently possess (see Yoshimura & MacWhinney, Reference Yoshimura and MacWhinney2010 that suggest that greater proficiency – serving as a proxy for increased L2 exposure – may enhance learners’ ability to weight and coordinate honorific cues effectively). Although our study does not address these possibilities directly, future research could explore these possibilities using diverse proficiency measures, to capture its multifaceted nature.
6. Conclusion
The findings of three studies collectively suggest that English-speaking learners of Korean may perceive and apply the subject honorific suffix as a general politeness marker rather than as a specific device for Korean subject honorification. This highlights the acquisitional challenges posed by low cue validity, frequent co-occurrence with other linguistic functions and socio–pragmatic complexity inherent in acquiring the independent function of subject honorification. Our results thus elucidate how cue competition and weighting shape L2 acquisitional trajectories, given the noisy representations involving L2 knowledge (Futrell & Gibson, Reference Futrell and Gibson2017; Tachihara & Goldberg, Reference Tachihara and Goldberg2020), as outlined in the Competition/Unified Model (MacWhinney, Reference MacWhinney, Robinson and Ellis2008, Reference MacWhinney, Gass and Mackey2013).
Although centered on a single target language – Korean – our findings have broader implications for L2 acquisition in lesser-studied languages. We have demonstrated how language-specific systems reveal general mechanisms of cue competition and differential weighting under noisy L2 representations, a perspective often obscured by the English-centric bias in the field. We show that learners reanalyzed a low-validity, socially salient morpheme (i.e., the subject-honorific suffix) as a global politeness cue, prioritizing semantic–pragmatic information over morphosyntactic constraints. These results support models that incorporate variable cue strength and frequent functional co-occurrence, predict task-sensitive processing differences, and permit pragmatic re-functionalization when structural cues are weak. Future work could calibrate cross-linguistic influence by testing parallel predictions across diverse L1 backgrounds, thereby enhancing generalizability to under-represented languages and learner populations.
As a final remark, we acknowledge that our study has limitations that warrant further investigation. We recruited only one learner group (English-speaking learners of Korean), and the findings may not generalize to other language pairs with respect to cross-linguistic influence. Therefore, future research should include learners from diverse L1 backgrounds that either share or differ in morphosyntactic features with Korean, alongside a larger sample size to further bolster statistical power, in order to enhance the generalizability of our findings. We believe that, while this extension would replicate the core findings of the present study, it can provide novel insights into how these typological similarities and differences between L1 and L2 influence the perception and use of this knowledge (cf. morphological congruency: Jiang et al., Reference Jiang, Novokshanova, Masuda and Wang2011; shared syntax: Hartsuiker & Bernolet, Reference Hartsuiker and Bernolet2017). In addition, precise investigation into the role of general cognitive capacities in acquiring this knowledge (cf. Cunnings, Reference Cunnings2017; McDonald, Reference McDonald2006), incorporating direct measurements of cognitive skills (e.g., working memory, inhibitory control) into data collection and analysis, could better illuminate the nature of (over-)generalization in L2 minds interfacing with cognitive resources available to them.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S1366728925100813.
Data availability statement
The data/code and all appendices in this study are available in the OSF repository at: https://osf.io/ngq7d/overview?view_only=990e5a93662d425797a398ee65981574.
Competing interests
The authors declare none.
Open access
