Earticle

This study aims to extract grammatical collocation patterns from the learner corpus constructed by the National Institute of the Korean Language (2015–2019) and to investigate the usage patterns of grammatical collocations among Korean language learners. Grammatical collocations are particularly interesting because they involve two or more linguistic units that function as a single, integrated entity and because these units include forms responsible for grammatical functions. In particular, grammatical collocations are not only a challenging aspect for foreign learners of Korean but also a crucial component in language instruction. Therefore, extracting grammatical collocations from a large-scale learner corpus and analyzing their characteristics is of utmost importance. To achieve this, this study extracted grammatical collocations according to proficiency levels from a morphologically analyzed learner corpus consisting of approximately 2.6 million word tokens and examined their characteristics. Furthermore, to highlight the distinctive features of grammatical collocations in the learner corpus, a comparative analysis was conducted with a native Korean corpus. Through this analysis, the study quantitatively and qualitatively examined the usage patterns of grammatical collocations among Korean learners based on their proficiency levels, while also explicitly identifying distinctions between learners' grammatical collocations and those of native speakers.

This study aims to identify and address the inherent political bias in ChatGPT by utilizing a sentiment analysis task. We set up representative figures from each political faction, asked chatgpt to write about the politicians using various prompts, and then sentimentally analyzed their outputs to determine the bias of ChatGPT. We found that ChatGPT is more positively biased toward liberal politicians in South Korea. We also found that ChatGPT's bias can be reduced by combining general narratives that encourage neutral writing or by refining the prompts with variables such as tone and writing style. This study provides important insights into the responsible use of AI and how to improve its bias.

Recently, the amount of newly coined words generated in Korean is vast, and the frequency of use in official language media such as the media, broadcasting, and books as well as everyday spoken language is gradually increasing. As the time spent in the Internet space increases, language for communication is created in various forms or its meaning changes to convey new information or values to members of society. In this study, the SNS corpus containing the rapidly changing use of language was analyzed. After selecting new word candidates by constructing a series of pipelines for extracting noun-type new words from the SNS corpus, characteristics and usage were analyzed. At this time, in the natural language processing pipeline that extracts new words, a pipeline including all the processes of rule-based learning using Mecab, unsupervised learning using Soynlp, and user dictionary addition using a correct morpheme analyzer was constructed to extract meaningful tokens. After completing the step of selecting new word candidates, 255 new words were collected. The proportion of sentences including the new word candidate group in the SNS data was 4.799%. Among them, the proportion of sentences in which words belonging to the top 10 appeared was 12.345%. Looking at the ratio of classifying the top 30 new words according to the word formation method, the word formation method that occupied the highest ratio was compound word-synthetic abbreviations (33.3%). The type/token ratio of sentence data including new words was 0.324. The type/token ratio of SNS data was 0.254. Since the type/token ratio of SNS data is lower, it can be said that ototoxicity is higher than that of sentences containing new words. When looking at the collocation relationship and usage of new words such as the initial constant word 'ㄹㅇ', the borrowed word '-특', and the meaning-expanded word '코인', various forms and syntactic uses could be found, and there were many collocations that reflected the social image at the time of data collection. Judging from this phenomenon, the characteristics of corpus, in which initials, borrowings, meeaning-expandings, and special characters are used among newly coined words, become incomplete when simply relying on a dictionary consisting of words or word lists, so a natural language processing dataset containing more diverse social meanings can be constructed by using usage data.

This study examines the semantic distinction between malda and beorida through corpus analysis. The non-substitutability of the two verbs is largely due to the syntactic and semantic constraints of malda, which are more restrictive than those of beorida. Despite similar conceptual meanings and syntactic combinations, differences emerge in morphological usage and modal nuances perceived by speakers. These findings suggest that malda and beorida form a challenging synonym pair for both teaching and learning, requiring careful semantic analysis. The identified constraints and differences may inform more effective materials for Korean language learners.

This study examined the semantic distinctions among the synonymous Korean adverbs gojak (‘고작’), gikkeot (‘기껏’), gyewu (‘겨우’), bulgwa (‘불과’), and gikkeot-haeya (‘기껏해야’) through a quantitative analysis based on co-occurrence data from the Sejong Corpus. Using hierarchical clustering and correspondence analysis, the study visualized the degree of semantic proximity among these adverbs.The hierarchical clustering results show that gikkeot and gikkeot-haeya form the closest semantic cluster, followed by gojak and bulgwa. In contrast, gyewu emerged as a semantically independent adverb, forming a distinct cluster. Correspondence analysis further confirmed these patterns by illustrating that gojak, gikkeot, and gikkeot-haeya are located near the origin and share similar directional vectors, indicating overlapping co-occurrence profiles. Meanwhile, bulgwa and gyewu are clearly separated in different quadrants of the plot, reflecting their distinct semantic and syntactic properties.By integrating co-occurrence patterns with statistical analysis, this study supplements intuition-based and dictionary-driven synonym classifications. The findings affirm that a corpus-based approach is effective in distinguishing subtle semantic differences among synonymous adverbs. Further research is needed to expand this analysis to a wider range of adverbs and to incorporate pragmatic and discourse-level factors into the investigation..

This study analyzes cultural biases in major large language models from the United States, South Korea, and China (GPT-4, CLOVA X, and Qwen1.5) through story generation tasks using culture-specific names. Morphological analysis of the generated stories revealed that all models exhibited certain cultural biases. GPT-4 did not show negative biases toward Korean and Chinese cultures but tended to prefer traditional and rural settings when describing these cultures. In contrast, CLOVA X and Qwen1.5, which are specialized for their respective national languages, portrayed their own cultures in modern and positive terms while using a relatively higher proportion of negative adjectives and unrealistic settings when describing Western contexts. These findings are significant because they go beyond the conventional focus on biases in Western-centric models toward non-Western contexts. They newly reveal that East Asian-based models can also exhibit similar biases when representing Western cultures. This research suggests that current language model has fundamental limitations in achieving cultural neutrality and highlights the importance of balanced learning and reflection of diverse cultural contexts as a crucial challenge in language model development.

The press not only delivers a wide range of news to the public but also plays a crucial role in shaping public opinion on various issues and situations. Depending on their interests, newspapers may interpret the same issue differently. One of the major topics consistently covered by the South Korean press is North Korea. Since the division of the Korean Peninsula, issues related to North Korea have remained a focal point in South Korean society. This study analyzes and discusses the lexical characteristics of North Korea-related news coverage according to the political orientation of newspapers. Politically charged high-frequency words were selected from both progressive and conservative newspapers. An analysis of the usage examples of these words reveals that progressive and conservative newspapers tend to view the same topic from differing perspectives when reporting on North Korea.

This study investigates the performance of a state-of-the-art Large Language Model(LLM) in classifying Korean neutral sentiment without additional fine-tuning and proposes an effective prompt design to improve neutral sentiment classification. To enhance the accuracy of neutral sentiment detection, we introduce a prompt based on Aspect-Based Sentiment Analysis(ABSA) and conduct a comparative evaluation using the proposed approach. The proposed prompt consists of a five-step procedure, including aspect identification and sentiment ratio–based classification, enabling more fine-grained sentiment reasoning. Experimental results demonstrate that the proposed prompt significantly improves classification accuracy when applied to the GPT-4-turbo model, thereby validating the effectiveness of prompt-based control for neutral sentiment analysis in Korean.

Developing AI-assisted application for pronunciation and speaking evaluation aids in helping students enhancing communicative competency. Recent advancement of speech technology and natural language processing makes it possible to build data-driven computer assisted language learning system with a limited capability of recognizing oral inputs. This paper overviews the needs and development such corpus-based computer assisted system of English learning and suggests a need for more sophisticated corpus construction for intermediate and advanced learners of English who are in need of enhancing their pronunciation and speaking competency.

This study analyzed the factors affecting the axiological semantic characteristics of Korean metaphors from the perspectives of cognitive linguistics and pragmatics. To this end, the related factors were classified into embodied characteristics, morality, and cultural relativistic characteristics. Finally, the face threatening aspects of each factor were analyzed through daily conversation corpus data. According to the results, many of the Korean metaphors with negative meanings were derived from proverbs or idiomatic expressions. In addition, the threat to face caused by metaphors containing cultural negativity was significant. In particular, the metaphorical expressions of animals or objects had a relatively high threat to face, as ontological negativity was added to idiomatic negativity. Nevertheless, the axiological semantic and embodied characteristics inherent in Korean metaphors were largely based on the universality suggested by previous studies. However, despite this universality, learners from other cultures appear to have difficulties in interpreting the meaning of Korean metaphors. Therefore, there is a need to prepare a systematic plan to minimize the pragmatic failure of Korean language learners