Hallucination in large language models (LLMs) remains a significant challenge for their safe deployment, particularly due to its potential to spread misinformation. Most existing solutions address this challenge by focusing on aligning the models with credible sources or by improving how models communicate their confidence (or lack thereof) in their outputs. While these measures may be effective in most contexts, they may fall short in scenarios requiring more nuanced approaches, especially in situations where access to accurate data is limited or determining credible sources is challenging. In this study, we take North Korea - a country characterised by an extreme lack of reliable sources and the prevalence of sensationalist falsehoods - as a case study. We explore and evaluate how some of the best-performing multilingual LLMs and specific language-based models generate information about North Korea in three languages spoken in countries with significant geo-political interests: English (United States, United Kingdom), Korean (South Korea), and Mandarin Chinese (China). Our findings reveal significant differences, suggesting that the choice of model and language can lead to vastly different understandings of North Korea, which has important implications given the global security challenges the country poses.

Soyoung Yang, Hojun Cho, Jiyoung Lee, Sohee Yoon, Edward Choi, Jaegul Choo, Won Ik Cho. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Phonetic features are indispensable in understanding the spoken language. Especially in Korean, which is wh-in-situ and head-final, the addressee of spoken language sometimes finds it hard to discern the speaker’s original intention if not provided with the sentence prosody. However, acoustic information may not be guaranteed for all spoken language processing, due to the difficulty of managing and computing speech data. This article suggests a corpus that aims to distinguish utterances with ambiguous intention from clear-cut ones, utilizing the prosodic ambiguity of the text input. In detail, the resulting classification system decides whether the given text input is one of fragment, statement, question, command, rhetorical question/command, or indecisive, taking into account the intonation-dependency of the text. Based on an intuitive understanding of the Korean language engaged in the data annotation, we construct a corpus with seven intention categories, train classification systems, and validate the utility of our dataset with quantitative and qualitative analyses.

Paraphrasing is often performed with less concern for controlled style conversion. Especially for questions and commands, style-variant paraphrasing can be crucial in tone and manner, which also matters with industrial applications such as dialog systems. In this paper, we attack this issue with a corpus construction scheme that simultaneously considers the core content and style of directives, namely intent and formality, for the Korean language. Utilizing manually generated natural language queries on six daily topics, we expand the corpus to formal and informal sentences by human rewriting and transferring. We verify the validity and industrial applicability of our approach by checking the adequate classification and inference performance that fit with conventional fine-tuning approaches, at the same time proposing a supervised formality transfer task.

Korean is a language with complex morphology that uses spaces at larger-than-word boundaries, unlike other East-Asian languages. While morpheme-based text generation can provide significant semantic advantages compared to commonly used character-level approaches, Korean morphological analyzers only provide a sequence of morpheme-level tokens, losing information in the tokenization process. Two crucial issues are the loss of spacing information and subcharacter level morpheme normalization, both of which make the tokenization result challenging to reconstruct the original input string, deterring the application to generative tasks. As this problem originates from the conventional scheme used when creating a POS tagging corpus, we propose an improvement to the existing scheme, which makes it friendlier to generative tasks. On top of that, we suggest a fully-automatic annotation of a corpus by leveraging public analyzers. We vote the surface and POS from the outcome and fill the sequence with the selected morphemes, yielding tokenization with a decent quality that incorporates space information. Our scheme is verified via an evaluation done on an external corpus, and subsequently, it is adapted to Korean Wikipedia to construct an open, permissive resource. We compare morphological analyzer performance trained on our corpus with existing methods, then perform an extrinsic evaluation on a downstream task.

Recently, a conventional neural decoder for speech codec has shown promising performance. However, it typically requires some prior knowledge of decoding such as bit allocation or dequantization scheme, which is not a universal solution for many different kinds of speech codecs. In order to address this limitation, we propose a neurally optimized decoder based on a generative model which can directly reconstruct the speech from the bitstream without a prior knowledge. The proposed decoder mainly consists of two components: 1) a dequantization model to group and dequantize related bits from the bitstream and 2) a generative model to restore the speech conditioned on the output of the dequantization model. Through experiments with mixed excitation linear prediction (MELP), Advanced multi-band excitation (AMBE), and SPEEX at around 2.4 kb/s, it is showed that the proposed model showed better performance in most of the objective and subjective evaluation compared to the conventional speech codecs.

For readability and disambiguation of the written text, appropriate word segmentation is recommended for documentation, and it also holds for the digitized texts. If the language is agglutinative while far from scriptio continua, for instance in the Korean language, the problem becomes more significant. However, some device users these days find it challenging to communicate via key stroking, not only for handicap but also for being unskilled. In this study, we propose a real-time assistive technology that utilizes an automatic word segmentation, designed for digital minorities who are not familiar with electronic typing. We propose a data-driven system trained upon a spoken Korean language corpus with various non-canonical expressions and dialects, guaranteeing the comprehension of contextual information. Through quantitative and qualitative comparison with other text processing toolkits, we show the reliability of the proposed system and its fit with colloquial and non-normalized texts, which fulfills the aim of supportive technology.

Recent voice activity detection (VAD) schemes have aimed at leveraging the decent neural architectures, but few were successful with applying the attention network due to its high reliance on the encoder-decoder framework. This has often let the built systems have a high dependency on the recurrent neural networks, which are costly and sometimes less context-sensitive considering the scale and property of acoustic frames. To cope with this issue with the self-attention mechanism and achieve a simple, powerful, and environment-robust VAD, we first adopt the self-attention architecture in building up the modules for voice detection and boosted prediction. Our model surpasses the previous neural architectures in view of low signal-to-ratio and noisy real-world scenarios, at the same time displaying the robustness regarding the noise types. We make the test labels on movie data publicly available for the fair competition and future progress.

Most speech-to-text (S2T) translation studies use English speech as a source, which makes it difficult for non-English speakers to take advantage of the S2T technologies. For some languages, this problem was tackled through corpus construction, but the farther linguistically from English or the more under-resourced, this deficiency and underrepresentedness becomes more significant. In this paper, we introduce kosp2e (read as `kospi'), a corpus that allows Korean speech to be translated into English text in an end-to-end manner. We adopt open license speech recognition corpus, translation corpus, and spoken language corpora to make our dataset freely available to the public, and check the performance through the pipeline and training-based approaches. Using pipeline and various end-to-end schemes, we obtain the highest BLEU of 21.3 and 18.0 for each based on the English hypothesis, validating the feasibility of our data. We plan to supplement annotations for other target languages through community contributions in the future.

Individuals interact with algorithms in various ways. Users even game and circumvent algorithms so as to achieve favorable outcomes. This study aims to come to an understanding of how various stakeholders interact with each other in tricking algorithms, with a focus towards online review communities. We employed a mixed-method approach in order to explore how and why users write machine non-translatable reviews as well as how those encrypted messages are perceived by those receiving them. We found that users are able to find tactics to trick the algorithms in order to avoid censoring, to mitigate interpersonal burden, to protect privacy, and to provide authentic information for enabling the formation of informative review communities. They apply several linguistic and social strategies in this regard. Furthermore, users perceive encrypted messages as both more trustworthy and authentic. Based on these findings, we discuss implications for online review community and content moderation algorithms.

In recent years, there has been a great deal of research in developing end-to-end speech recognition models, which enable simplifying the traditional pipeline and achieving promising results. Despite their remarkable performance improvements, end-to-end models typically require expensive computational cost to show successful performance. To reduce this computational burden, knowledge distillation (KD), which is a popular model compression method, has been used to transfer knowledge from a deep and complex model (teacher) to a shallower and simpler model (student). Previous KD approaches have commonly designed the architecture of the student by reducing the width per layer or the number of layers of the teacher. This structural reduction scheme might limit the flexibility of model selection since the student model structure should be similar to that of the given teacher. To cope with this limitation, we propose a KD method for end-to-end speech recognition, namely TutorNet, that applies KD techniques across different types of neural networks at the hidden representation-level as well as the output-level. For concrete realizations, we firstly apply representation-level knowledge distillation (RKD) during the initialization step, and then apply the softmax-level knowledge distillation (SKD) combined with the original task learning. When the student is trained with RKD, we make use of frame weighting that points out the frames to which the teacher pays more attention. Through a number of experiments, it is verified that TutorNet not only distills the knowledge between networks with different topologies but also significantly contributes to improving the performance of the distilled student.

Speech is one of the most effective means of communication and is full of information that helps the transmission of utterer's thoughts. However, mainly due to the cumbersome processing of acoustic features, phoneme or word posterior probability has frequently been discarded in understanding the natural language. Thus, some recent spoken language understanding (SLU) modules have utilized end-to-end structures that preserve the uncertainty information. This further reduces the propagation of speech recognition error and guarantees computational efficiency. We claim that in this process, the speech comprehension can benefit from the inference of massive pre-trained language models (LMs). We transfer the knowledge from a concrete Transformer-based text LM to an SLU module which can face a data shortage, based on recent cross-modal distillation methodologies. We demonstrate the validity of our proposal upon the performance on Fluent Speech Command, an English SLU benchmark. Thereby, we experimentally verify our hypothesis that the knowledge could be shared from the top layer of the LM to a fully speech-based module, in which the abstracted speech is expected to meet the semantic representation.

Assessing the similarity of sentences and detecting paraphrases is an essential task both in theory and practice, but achieving a reliable dataset requires high resource. In this paper, we propose a discourse component-based paraphrase generation for the directive utterances, which is efficient in terms of human-aided construction and content preservation. All discourse components are expressed in natural language phrases, and the phrases are created considering both speech act and topic so that the controlled construction of the sentence similarity dataset is available. Here, we investigate the validity of our scheme using the Korean language, a language with diverse paraphrasing due to frequent subject drop and scramblings. With 1,000 intent argument phrases and thus generated 10,000 utterances, we make up a sentence similarity dataset of practically sufficient size. It contains five sentence pair types, including paraphrase, and displays a total volume of about 550K. To emphasize the utility of the scheme and dataset, we measure the similarity matching performance via conventional natural language inference models, also suggesting the multi-lingual extensibility.