This paper reviews and critically discusses some widely believed but not entirely evidenced arguments for the so-called Internally-Headed Relative Clauses (IHRCs) in Korean. We deal with two highly controversial issues: the definition issue and the validation issue of the most believed grammatical peculiarities of IHRCs in Korean. We first deal with the previous attempts to define IHRCs in Korean, showing that they fail to differentiate IHRCs from other relative clauses. Furthermore, we show that, contrary to the definition of the relative clauses, IHRCs are neither subordinate clauses nor nominalized sentences in Korean. In addition, we show that the overtness of the pivot element cannot be maintained. We then critically discuss the issue of whether the mostly assumed grammatical peculiarities of the Korean IHRCs get empirically supported by a set of new data. Finally, we dispute widely assumed properties of IHRCs in Korean such as gaplessness, unbounded dependencies, the identity of truth-conditional meaning between IHRCs and Externally-Headed Relative Clauses (EHRCs), and non-restrictive interpretation with maximality effects. .

Grapheme-To-Phoneme (G2P) converts an input letter sequence into its corresponding pronunciation. Several rule-based G2P models are available for Korean, but their performance in view of individual phonological rules in Korean is not well-investigated. We establish phonological rules to be reflected in a G2P model for Korean, and test the model performance of two most popular G2P models of Korean, i.e., g2pk and KoG2P, regarding their performance on each of these rules. We created a golden corpus to evaluate the performance of the current G2P models based on manual phonetic transcription. We then measured the performance of two G2P models, and identified the phonological rules in which the models show relative success or failure in deriving the correct output. We implemented additional phonological rules such as h-deletion, ui-variation, consonant place assimilation, and restructured the ordering of rules such as h/th-neutralization in the model. We show that our revised model makes a substantial improvement on model performance. Further, we argue that the major limitation of the current rule-based approaches to G2P is in its binary approach to phonological rules and lack of information about prosodic boundaries. We propose that a rule-based G2P system should reflect the stochastic nature of phonological processes informed by existing research on the gradient nature of phonological rule application as a function of factors such as lexical frequency and speech register.

Through examining the successful building experience of the well-known Korean and foreign language NLP benchmarks and closely comparing their task list compositions, the study intends to provide a novel organizing scheme for the task list in the future-oriented Korean benchmarks. When creating benchmarks for a low-resource language like Korean, as opposed to English, a worldwide language, there may be a variety of obstacles, from procuring language resources to identifying important demands in a real-world setting. To figure out a more efficient and economical construction plan for the next-generation Korean benchmark, it is of utmost necessity to pay special attention to the organization of the task list, which serves as the basis of the benchmark, Consequently, in this study, we propose a new organization plan for the task list of the Korean benchmarks based on an insightful investigation of the various task list organization methods in the previous NLP benchmarks. Additionally, we suggest a step-by-step development roadmap for Korean benchmarks through the prioritization of the various tasks, incorporating the comprehensive consideration of the commonality of the evaluation system for the AI's language ability and the specificity of the development of the Korean AI model.

This study aimed to develop the Korean-version LexTALE and examine its effectiveness in estimating Korean learners' proficiency. 41 advanced and intermediate Korean learners participated in a web-based lexical decision task in which 120 Korean words and nonwords were tested. They also took part in a reading comprehension task and a cloze task. The results from a series of logistic mixed-effect models revealed that first, the LexTALE-Kr accuracy significantly differed between language groups in a way that Korean natives obtained higher scores than Korean learners, and advanced Korean learners gained higher scores than intermediate Korean learners. Second, the effect of lexical frequency on lexical decision accuracy was significant, and its effect was the biggest in advanced Korean learners. Third, the scores obtained from the reading and cloze tasks significantly accounted for the accuracy variances of Korean learners. Taken together, our results supported the claim that the simple and low-cost online test evaluating learners' lexical knowledge is adequate to estimate the goodness of learners' proficiency as much as complex and high-cost comprehensive examinations like TOPIK.

The purpose of this study is to extract unethical use cases from the Korean dictionary to determine whether they are ethical and to find the right way to present ethical use cases. In this study, the problem of misclassification and undetected ethical automatic classifier of the Korean dictionary example was analyzed through the crowd worker and the ethical automatic classifier 'Safety Check(Korv0.5). This study argued that cases with long-established unethical bias should be reviewed again through unethical cases in the Korean dictionary. Ethical information should be structured according to the purpose of the dictionary, but unethical use cases should be given additional attachment information, or in the case of new words, ethical regulations are needed in the guidelines for preparing examples. The unethical case analysis data reviewed in this study were disclosed to help improve the model through balanced data learning in subsequent studies.

In this paper, I make a formal semantic description of Korean conjunctive sentences which are headed by connective ending '-killay'. Through a critique of the previous formal semantic analysis of '-killay' by Yeom(2012), I argue that '-killay' has a direct evidential meaning which is interpreted as a kind of epistemic modal meaning, every type of '-killay' conjunctive sentences has presuppositions that the consequent proposition is true and the consequent situation is stage-level, and the time of acquiring direct evidence about the antecedent situation precedes the time of the consequent situation which in turn precedes or overlaps the utterance time. The lexical semantic definition containing all these properties of '-killay' is shown to make complete the semantic composition procedures of '-killay' conjunctive sentences. (Yonsei University).

This study evaluated whether an English language models using artificial intelligence represent world knowledge, focusing on the backshift phenomenon. Backshift refers to the past tense being used in indirect speech, with direct speech being in the present tense. This study argued that language models capture grammatical phenomena that interact with world knowledge and can represent knowledge beyond mere grammatical knowledge. This study used BERT and mBERT models to measure the surprisals of verbs in indirect speech. Surprisal is a measurement that increase in proportion to the difficulty of language processing. Experimental results demonstrated that BERT and mBERT models were sensitive to the backshift phenomenon, revealing that the artificial intelligence language model captured backshift in indirect speech. As backshift requires an understanding of truth, knowledge, and commonsense, these results indicated that the language model understood world knowledge. (Korea University).

This squib investigates the semantic value, discourse context, syntactic structure and prosody of a polar question with a focus (F-polQ) in Korean, e.g., JONF-i oassni? 'Did JOHNF come?', which has not been paid much attention to. First we observe that in generating an alternative set, an F-polQ goes with a focused declarative sentence (F-Decl) on the one hand, and in that its ordinary semantic value is the set of positive and negative propositions, {p, ￢p}, it goes with an ordinary polar question on the other. Based on this, this study shows that i) the alternative set of propositions generated by an F-polQ refers to the presupposed wh-question, just like an F-Decl does, and that ii) whereas the wh-Question that provides a proper context for an occurrence of a F-Decl originates from the hearer part, forming a question-answer dialogue, the wh-Question generated by a F-polQ, indicated by the prosody of L- phrase accent, originates from the speaker part. Additionally, it is suggested that the rising boundary tone H% indicates the source of the wh-question presupposed by the F-polQ is the speaker herself.