Songs feel emotionally different to listeners depending on their lyrical contents, even when melodies are similar. Accordingly, when using features related to melody, like tempo, rhythm, tune, and musical note, it is difficult to classify emotions accurately through the existing music emotion classification methods. This paper therefore proposes a method for lyrics-based emotion classification using feature selection by partial syntactic analysis. Based on the existing emotion ontology, four kinds of syntactic analysis rules were applied to extract emotion features from lyrics. The precision and recall rates of the emotion feature extraction were 73% and 70%, respectively. The extracted emotion features along with the NB, HMM, and SVM machine learning methods were used, showing a maximum accuracy rate of 58.8%.

As the volume and diversity of scientific resources grows, trend detection and analysis have become much more important issues. A variety of trend detection, characterization, evaluation and visualization methodologies have been introduced for various application domains. In this paper, we consider detection of temporal trends for topics using feature selection and extraction of additional information from the subtopics of them, for the Global Trends Briefing (GTB) dataset. Thus, we propose a novel trend detection method using feature selection based on the Improved Gini-Index (I-GI) algorithm, which can obtain representative features for given topics. Second, with those features, we extract subtopics for the topics and visualize temporal/emerging/upward/ downward trends with them. Third, utilizing the relations among the subtopics, we obtain relevant documents and seed sentences that co-occur with the upper features for the topic. In addition, we can extract information to forecast future trends relevant to the issues: for example, financial market or emerging technology. In the experimental results, we could obtain good representative features, more specific trends for the topics, and additional useful information.

Prediction of the prosodic phrase boundary is a potent influence on the performance of speech recognition and voice synthesis systems. We propose a statistical approach using efficient learning features for the natural prediction of the Korean prosodic phrase boundary. These new features reflect factors that affect the generation of the prosodic phrase boundary better than existing learning features. Notably, moreover, learning features that are extracted according to the hand-crafted prosodic phrase prediction rule impart higher accuracy. We evaluated the new learning features in terms of their efficiency in predicting the prosodic phrase boundary, using CRFs (conditional random fields). The results were 84.63% accuracy for three levels and 80.14% accuracy for six levels.

Despite the increasing need for accuracy, current text-to-speech (TTS) systems are still poor at generating the correct pronunciation of Arabic numerals due to their high ambiguity and various interpretations. In this paper, we propose a mini-transliteration system for Arabic-numeral expressions, which can efficiently and correctly convert Arabic numeral expressions found in Korean text into phonemes for embedded TTS systems. For the purpose of building grapheme-to-phoneme rules, we deduced the components of ANEs, and investigated their pattern and arithmetic features based on the analyzed corpus. A word sense disambiguation based on lexical hierarchies in KorLex 1.0 was developed to resolve ambiguities caused by the homographic components of the ANEs. Our system minimized the amount of memory used by 1) separating the morphological analysis module from the transliteration system, 2) compacting the lexicon size, and 3) removing named entities. It reduced the process time dramatically without any serious loss of accuracy, and showed an accuracy of 97.2%-98.3%, which was 21.4%-22.5% higher than that of the baseline, and 5.5%-19.5% higher than current commercial Korean TTS systems

We have established a preprocessing method for determining the meaningfulness of a table to allow for information extraction from tables on the Internet. A table offers a preeminent clue in text mining because it contains meaningful data displayed in rows and columns. However, tables are used on the Internet for both knowledge structuring and document design. Therefore, we were interested in determining whether or not a table has meaningfulness that is related to the structural information provided at the abstraction level of the table head. Accordingly, we: 1) investigated the types of tables present in HTML documents, 2) established the features that distinguished meaningful tables from others, 3) constructed a training data set using the established features after having filtered any obvious decorative tables, and 4) constructed a classification model using a decision tree. Based on these features, we set up heuristics for table head extraction from meaningful tables, and obtained an F-measure of 95.0 percent in distinguishing meaningful tables from decorative tables and an accuracy of 82.1 percent in extracting the table head from the meaningful tables.

Despite of much work on TTS technologies and several TTS systems customized for Korean, current TTS systems output many errors in transliterating non-alphabetic symbols such as Arabic numerals and text symbols. This paper proposes TLAN (Transliteration Learner for Arabic-Numeral expressions) which can efficiently disambiguate the reading and meaning of Arabic Numeral Expressions (ANEs) in texts by using a decision tree. For the purpose of analyzing and learning data, three phases of learning elements were suggested: patterns of Arabic numerals combined with text symbols, contextual features and heuristic information were classified according to the senses and sounds of ANEs. Our corpus was made up of news articles issued from January 1, 2000 to December 31, 2001 from 10 major newspapers in Korea. By learning the three phases of learning elements, the model shows 97.38% and 97.28% accuracies for the training set and the test set, respectively.

The main aim of this work is to make improvements to Mirine 2.2, a Korean grammar-checker that takes Korean language properties into account and to develop a system that satisfies user expectations. Particular attention is given to processing Korean texts that may contain users' grammatical and practical errors. In order to treat these errors efficiently, the system includes methods based on asymmetric relations from which partial parsing and the potential governing relationship are derived, implying the starting point for checking, the direction for parsing, and the limits of the scope for parsing.To organize partial parsing efficiently, the system requires an appropriate knowledge base. As its essential prerequisites, this study (a) considers the factors from which the various error types encountered while parsing various Korean texts arise; (b) extracts general patterns from the linguistic or extra-linguistic factors obtained in this manner; and (c) demonstrates how the system, based on a linguistic analysis, procures an adequate knowledge base for partial parsing to satisfy end-users. Mirine 2.2 achieves an F-measure of about 0.80 in detecting unknown erroneous words; an F-measure of about 0.98 when not considering unknown words; and 98.94% precision in correcting erroneous words.

The approach in this paper focuses on the contextual postprocessing by selecting the most feasible word from multiple output strings of an OCR system. The correction is applied only when the selection fails. The selected word is confirmed by the collocation between the word and the adjacent words. The five functions applied in the system are (1) to select a word from candidate words, (2) to correct candidate words using a confusion matrix of syllables, (3) to combine two substrings to a word that spans two lines, (4) to guess unknown nouns, and (5) to confirm a selected word by the contextual information of adjacent words. To improve speed, we use syllable di-grams and viable-prefixes of Korean words. The experimental result shows that the two heuristics speed up the system more than 1,000 times in worst case. Our system improves the word recognition rate of the OCR system from 90.50% to 94.72%.