Natural Language Inference (NLI) is a fundamental task in Natural Language Processing (NLP).However, adapting NLI models to new domains remains challenging due to the high cost of collecting domain-specific training data.While prior work proposed 15 sentence transformation rules to automate training data generation, these rules do not sufficiently capture the diversity of natural language.We propose a novel framework that combines Out-of-Distribution (OOD) detection and BERT-based clustering to identify premise-hypothesis pairs in the SNLI dataset that are not covered by existing rules and discover four new transformation rules from them.Using these rules with Chain-of-Thought (CoT) prompting and Large Language Models (LLMs), we generate high-quality training data and augment the SNLI dataset.Our method yields consistent performance improvements across dataset sizes, achieving +0.85%p accuracy on 2k samples and +0.15%p on 550k samples.Furthermore, a distribution-aware augmentation strategy enhances performance across all scales.Beyond manual explanations, we extend our framework to automatically-generated explanations (CoT-Ex), demonstrating that they provide a scalable alternative to human-written explanations and enable reliable rule discovery.

Machine reading comprehension question answering (MRC-QA) is the task of understanding the context of a given passage to find a correct answer within it. A passage is composed of several sentences; therefore, the length of the input sentence becomes longer, leading to diminished performance. In this article, we propose S 3 -NET, which adds sentence-based encoding to solve this problem. S 3 -NET, which is based on a simple recurrent unit architecture, is a deep learning model that solves the MRC-QA by applying matching network to sentence-level encoding. In addition, S 3 -NET utilizes self-matching networks to compute attention weight for its own recurrent neural network sequences. We perform MRC-QA for the SQuAD dataset of English and MindsMRC dataset of Korean. The experimental results show that for SQuAD, the S 3 -NET model proposed in this article produces 71.91% and 74.12% exact match and 81.02% and 82.34% F1 in single and ensemble models, respectively, and for MindsMRC, our model achieves 69.43% and 71.28% exact match and 81.53% and 82.77% F1 in single and ensemble models, respectively.

Dependency parsing is a fundamental problem in natural language processing. We introduce a novel dependency-parsing framework called head-pointing--based dependency parsing . In this framework, we cast the Korean dependency parsing problem as a statistical head-pointing and arc-labeling problem. To address this problem, a novel neural network called the multitask pointer network is devised for a neural sequential head-pointing and type-labeling architecture. Our approach does not require any handcrafted features or language-specific rules to parse dependency. Furthermore, it achieves state-of-the-art performance for Korean dependency parsing.

In this article, we introduce a novel neural architecture for the end-to-end Korean Part-of-Speech (POS) tagging problem. To address the problem, we extend the present recurrent neural network-based sequence-to-sequence models to deal with the key challenges in this task: rare word generation and POS tagging. To overcome these issues, Input-Feeding and Copying mechanism are adopted. Although our approach does not require any manual features or preprocessed pattern matching dictionaries, our best single model achieves an F-score of 97.08. This is competitive with the current state-of-the-art model (F-score 98.03), which requires extensive manual feature processing.

Most studies on statistical Korean word spacing do not utilize the information provided by the input sentence and assume that it was completely concatenated.This makes the word spacer ignore the correct spaced parts of the input sentence and erroneously alter them.To overcome such limit, this paper proposes a structural SVM-based Korean word spacing method that can utilize the space information of the input sentence.The experiment on sentences with 10% spacing errors showed that our method achieved 96.81% F-score, while the basic structural SVM method only achieved 92.53% F-score.The more the input sentence was correctly spaced, the more accurately our method performed.

In this paper, we describe a named entity recognition using a modified Pegasos algorithm for structural SVMs. We show the modified Pegasos algorithm significantly outperformed CRFs and the training time for the modified Pegasos algorithm is reduced 17-26 times compared to CRFs.

In this paper, we describe a fixed-threshold sequential minimal optimization (FSMO) for a joint constraint learning algorithm of structural classification SVM problems. Because FSMO uses the fact that the joint constraint formulation of structural SVM has b=0, FSMO breaks down the quadratic programming (QP) problems of structural SVM into a series of smallest QP problems, each involving only one variable. By using only one variable, FSMO is advantageous in that each QP sub-problem does not need subset selection.

No abstract available.

Abstract In text categorization tasks, classification on some class hierarchies has better results than in cases without the hierarchy. Currently, because a large number of documents are divided into several subgroups in a hierarchy, we can appropriately use a hierarchical classification method. However, we have no systematic method to build a hierarchical classification system that performs well with large collections of practical data. In this article, we introduce a new evaluation scheme for internal node classifiers, which can be used effectively to develop a hierarchical classification system. We also show that our method for constructing the hierarchical classification system is very effective, especially for the task of constructing classifiers applied to hierarchy tree with a lot of levels.

This paper presents an Information Extraction (IE) approach for spoken language understanding. The goal in IE is to find proper values for pre-defined slots of given templates. IE for spoken language understanding proposes a concept spotting approach for spoken language because IE approach is interested in only pre-defined concept slots. In spite of this partial understanding, we can acquire necessary information for an application from the values of pre-defined slots because the slots are properly designed for speech understanding in a specific domain. Spoken language has so many recognition errors especially in a poor environment so it is more difficult to understand than textual language. Considering this fact, we attempt to understand the languages by concentrating on the specified information. In experiments on the car navigation domain, F-measure for concept spotting for textual input (WER 0%) and spoken input (WER 39%) are 96.33 % and 78.30 % respectively. 1.