Earticle

This study investigates the consistency and agreement between scores generated by human raters and an artificial intelligence (AI)-based automated scoring system in the context of high school English writing performance assessments. To evaluate the pedagogical validity of AI-embedded scoring tools, 140 student essays were independently assessed by three raters and a commercial AI system using an analytical rubric. Statistical analysis utilizing Intraclass Correlation Coefficients was conducted to measure inter-rater reliability, the internal consistency of the AI system, and the degree of human-AI agreement. The results indicate that while human raters maintained moderate-to-high agreement and the AI system demonstrated high stability across iterated measures, the agreement between the two varied significantly across construct domains. Notably, a consistently low correlation was identified in the domain of Expression of Opinion. The findings suggest that while AI systems reliably process structured linguistic features such as lexis, grammar and textual coherence, they may operationalize different construct interpretations than human raters when evaluating context-sensitive competencies such as discursive meaning and rhetorical strategies. The paper concludes by underscoring the need for collaborative human-AI assessment models and further validation of construct validity in automated evaluation.

This study proposes and examines strategies for integrating Artificial Intelligence(AI) into an LMS-based course, ｢Research in Korean Language Education｣, to enhance learners’ text summarization and question-answer generation skills. A three-stage instructional model was implemented with academic-purpose learners, followed by a survey to evaluate learning outcomes and perceptions. Throughout the course, AI served as a comparative tool against learners’ self-generated ideas. The findings yield three pedagogical implications for AI integration. First, in text summarization, AI-generated outputs should be critically compared with learners’ summaries to foster discovery learning. Second, course design must systematically align summarization and question-answer tasks with the enhancement of four core comprehension skills. Third, instructional emphasis should fall on constructive feedback and discussion rather than output generation alone. This study demonstrates how AI-assisted instruction can promote discovery learning and heuristic thinking among learners of Korean for academic purposes.

This paper investigates the so-called ‘selected object’ resultatives in Korean. Adopting the adjunct analysis of ‘selected object’ resultatives, this study aims to reveal the structure of the result constituent, focusing on whether or not the DP object of the verb is syntactically associated with the secondary predicate. In this study, I present novel evidence—concerning verbs of implicitly created objects and verbs derived from result nominals—that supports the ‘nonclausal’ analysis of ‘selected object’ resultatives. Based on the parallel behaviors of the new and traditional ‘selected object’ resultatives, I argue that the Res head in the result constituent (i.e., ResP) directly selects for the resultative predicate projection. In other words, there is no syntactic position inside the ResP for the putative logical subject co-indexed with the verbal object. I extend the current ‘nonclausal’ analysis to resultatives built on a subset of intransitive verbs (i.e., Levin(1993)’s ‘dress’ verbs), thereby calling into question the ‘selected’/‘nonselected’ object dichotomy. The results of this study suggest that Korean resultatives are distinguished in terms of the implicational relationship between (non)clausality of the ResP and the vP’s result semantics, rather than whether or not the verbal object is selected by the primary predicate.

The ultimate goal of this paper is to provide an in-depth analysis of the interpretations induced by conjunct NPs and Korean anaphors. One important observation is that, in local contexts, caki ‘self’ can function both as a pronoun and as an anaphor, whereas with conjunct NPs it behaves as a pronoun. Furthermore, non-local caki ‘self’ and caki-casin ‘self-self’ denotes a singular entity whose referent must be singular. This property enables them to yield both distributive and bound-variable readings with respect to conjunct NPs and quantified phrases (QPs). When the plural morpheme tul ‘PL’ is attached to caki ‘self’, it triggers the interpretation “x and others.” By contrast, when tul ‘PL’ is attached to caki-casin ‘self-self’, this interpretation does not arise. Finally, this paper argues that the distributive and collective readings induced by Korean anaphors can be explained in terms of Chomsky’s(2019a, 2019b) operations of Merge and Transfer.

Chomsky(2024) proposes Box System(BS) while eliminating the segregation of A/A’- positions and A- and A’-movement. He applies this system to several constructions in the paper. However, he does not address how passive interrogative constructions are derived within the BS framework. The purpose of the present study is to reveal how passive interrogative constructions are formed within the BS framework, centering around a wh-object in single transitive constructions and a wh-indirect object in double-object constructions(DOC). To this end, this study (i) examines the motivation for eliminating the segregation of A/A’-positions and A- and A’-movement, (ii) demonstrates that Chomsky’s(2024) BS is more suitable than Kitahara and Seely’s (2025) version of BS, (iii) shows in detail how wh-objects are externalized within the Chomsky’s(2024) BS framework, (iv) demonstrates that passivized wh-subjects undergo boxing at the inner phase of v*P during the course of derivation, and (v) addresses why a wh-direct object in DOC cannot be passivized.

This corpus-based study examines whether written Korean exhibits a syllable-internal sequential asymmetry and, if so, whether the pattern is analogous to the probabilistic C1V bias reported for spoken Korean(Park, 2023). Using a dataset of mono- and disyllabic nouns extracted from the Written Corpus, this study measured contingency between adjacent graphemes within C1VC2 syllables with phi coefficients and analyzed the distributions of the C1V and VC2 sequences. The results provide no distributional evidence for either a C1V bias or a VC2 bias, suggesting that the C1V bias observed in spoken Korean may not be modality-neutral but may be sensitive to the distributional ecologies of written and spoken input. This study extends the usage-based perspective to the written modality by treating written language as a potential input domain for distributional learning and contributes to a more refined account of structural emergence.