• Language and Information
• /
• v.2 no.1
• /
• pp.79-105
• /
• 1998

This paper aims to describe a German-Korean machine translation system for nominal phrases. Besides, we have two subgoals. First, we are going to revea linguistic differences between two languages and propose a language-informational method fo overcome the differences. The method is based on an integrated model of translation knowledge, efficient information structure, and concordance selection. Then, we will show the statistical results about translation experiment and its evaluation as an evidence for the adequacy of our linguistic method and translation system itself.

• KIISE Transactions on Computing Practices
• /
• v.23 no.7
• /
• pp.446-451
• /
• 2017

The statistical context-sensitive spelling correction technique in this thesis is based upon Shannon's noisy channel model. The interpolation method is used for the improvement of the correction method proposed in the paper, and the general interpolation method is to fill the middle value of the probability by (N-1)-gram and (N-2)-gram. This method is based upon the same statistical corpus. In the proposed method, interpolation is performed using the frequency information between the statistical corpus and the correction document. The advantages of using frequency of correction documents are twofold. First, the probability of the coined word existing only in the correction document can be obtained. Second, even if there are two correction candidates with ambiguous probability values, the ambiguity is solved by correcting them by referring to the correction document. The method proposed in this thesis showed better precision and recall than the existing correction model.

• ETRI Journal
• /
• v.46 no.1
• /
• pp.48-58
• /
• 2024

This paper presents the development of language tutoring systems for nonnative speakers by leveraging advanced end-to-end automatic speech recognition (ASR) and proficiency evaluation. Given the frequent errors in non-native speech, high-performance spontaneous speech recognition must be applied. Our systems accurately evaluate pronunciation and speaking fluency and provide feedback on errors by relying on precise transcriptions. End-to-end ASR is implemented and enhanced by using diverse non-native speaker speech data for model training. For performance enhancement, we combine semisupervised and transfer learning techniques using labeled and unlabeled speech data. Automatic proficiency evaluation is performed by a model trained to maximize the statistical correlation between the fluency score manually determined by a human expert and a calculated fluency score. We developed an English tutoring system for Korean elementary students called EBS AI Peng-Talk and a Korean tutoring system for foreigners called KSI Korean AI Tutor. Both systems were deployed by South Korean government agencies.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.11
• /
• pp.4184-4202
• /
• 2021

Text steganography combined with natural language generation has become increasingly popular. The existing methods usually embed secret information in the generated word by controlling the sampling in the process of text generation. A candidate pool will be constructed by greedy strategy, and only the words with high probability will be encoded, which damages the statistical law of the texts and seriously affects the security of steganography. In order to reduce the influence of the candidate pool on the statistical imperceptibility of steganography, we propose a steganography method based on a new sampling strategy. Instead of just consisting of words with high probability, we select words with relatively small difference from the actual sample of the language model to build a candidate pool, thus keeping consistency with the probability distribution of the language model. What's more, we encode the candidate words according to their probability similarity with the target word, which can further maintain the probability distribution. Experimental results show that the proposed method can outperform the state-of-the-art steganographic methods in terms of security performance.

• Journal of Information Processing Systems
• /
• v.17 no.1
• /
• pp.203-212
• /
• 2021

To eliminate ambiguities in the existing methods to simplify Chinese pronunciation learning, we propose a model that can predict the pronunciation of Chinese characters automatically. The proposed model relies on a statistical machine translation (SMT) framework. In particular, we consider the components of Chinese characters as the basic unit and consider the pronunciation prediction as a machine translation procedure (the component sequence as a source sentence, the pronunciation, pinyin, as a target sentence). In addition to traditional features such as the bidirectional word translation and the n-gram language model, we also implement a component similarity feature to overcome some typos during practical use. We incorporate these features into a log-linear model. The experimental results show that our approach significantly outperforms other baseline models.

• Journal of Advanced Marine Engineering and Technology
• /
• v.36 no.5
• /
• pp.683-693
• /
• 2012

Recently, statistical machine translation (SMT) has received many attention with ease of its implementation and maintenance. The goal of our works is to build bidirectional Korean-Japanese SMT system by using MOSES [1] system. We use Korean-Japanese bilingual corpus which is aligned per sentence to train the translation model and use a large raw corpus in each language to train each language model. The proposed system shows results comparable to those of a rule-based machine translation system. Most of errors are caused by noises occurred in each processing stage.

• International Journal of Computer Science & Network Security
• /
• v.23 no.9
• /
• pp.120-128
• /
• 2023

In this work, we have presented ATSA, a hierarchical attention deep learning model for Arabic sentiment analysis. ATSA was proposed by addressing several challenges and limitations that arise when applying the classical models to perform opinion mining in Arabic. Arabic-specific challenges including the morphological complexity and language sparsity were addressed by modeling semantic composition at the Arabic morphological analysis after performing tokenization. ATSA proposed to perform phrase-chunks sentiment embedding to provide a broader set of features that cover syntactic, semantic, and sentiment information. We used phrase structure parser to generate syntactic parse trees that are used as a reference for ATSA. This allowed modeling semantic and sentiment composition following the natural order in which words and phrase-chunks are combined in a sentence. The proposed model was evaluated on three Arabic corpora that correspond to different genres (newswire, online comments, and tweets) and different writing styles (MSA and dialectal Arabic). Experiments showed that each of the proposed contributions in ATSA was able to achieve significant improvement. The combination of all contributions, which makes up for the complete ATSA model, was able to improve the classification accuracy by 3% and 2% on Tweets and Hotel reviews datasets, respectively, compared to the existing models.

• Annual Conference on Human and Language Technology
• /
• 1996.10a
• /
• pp.21-31
• /
• 1996

복합명사는 한국어에서 가장 빈번하게 나타나는 색인어의 한 형태로서, 영어권 중심의 정보검색 모델로는 다루기가 어려운 언어 현상의 하나이다. 복합명사는 2개 이상의 단일어들의 조합으로 이루어져 있고, 그 형태 또한 여러 가지로 나타나기 때문에 색인과 검색의 큰 문제로 여겨져 왔다. 본 논문에서는 복합명사의 어휘적 정보를 단위명사들의 통계적 행태(statistical behavior)에 기반 하여 자동 획득하고, 이러한 어휘적 정보를 검색에 적용하는 모텔을 제시하고자 한다. 본 방법은 색인시의 복합명사 인식의 어려움과 검색시의 형태의 다양성을 극복하는 모델로서 한국어를 포함한 동양권의 언어적 특징을 고려한 모델이다.

• Journal of Information Processing Systems
• /
• v.18 no.5
• /
• pp.650-664
• /
• 2022

Mobile applications can be easily downloaded and installed via markets. However, malware and malicious applications containing unwanted advertisements exist in these application markets. Therefore, smartphone users install applications with reference to the application review to avoid such malicious applications. An application review typically comprises contents for evaluation; however, a false review with a specific purpose can be included. Such false reviews are known as fake reviews, and they can be generated using artificial intelligence (AI)-based text-generating models. Recently, AI-based text-generating models have been developed rapidly and demonstrate high-quality generated texts. Herein, we analyze the features of fake reviews generated from Generative Pre-Training-2 (GPT-2), an AI-based text-generating model and create a model to detect those fake reviews. First, we collect a real human-written application review from Kaggle. Subsequently, we identify features of the fake review using natural language processing and statistical analysis. Next, we generate fake review detection models using five types of machine-learning models trained using identified features. In terms of the performances of the fake review detection models, we achieved average F1-scores of 0.738, 0.723, and 0.730 for the fake review, real review, and overall classifications, respectively.

• Speech Sciences
• /
• v.11 no.4
• /
• pp.43-52
• /
• 2004

Model parameters in HMM based speech recognition systems are normally estimated using Maximum Likelihood Estimation(MLE). The MLE method is based mainly on the principle of statistical data fitting in terms of increasing the HMM likelihood. The optimality of this training criterion is conditioned on the availability of infinite amount of training data and the correct choice of model. However, in practice, neither of these conditions is satisfied. In this paper, we propose a training algorithm, MCE(Minimum Classification Error), to improve the performance of a speech recognizer detecting mispronunciation of a foreign language. During the conventional MLE(Maximum Likelihood Estimation) training, the model parameters are adjusted to increase the likelihood of the word strings corresponding to the training utterances without taking account of the probability of other possible word strings. In contrast to MLE, the MCE training scheme takes account of possible competing word hypotheses and tries to reduce the probability of incorrect hypotheses. The discriminant training method using MCE shows better recognition results than the MLE method does.