Search

To generate a proper Korean predicate, a natural modal expression is the most important factor for a machine translation (MT) system. Tense, aspect, mood, negation, and voice are the major constituents related to modal expression. The linguistic encoding of a modal expression is quite different between Chinese and Korean in terms of linguistic typology and genealogy. In this paper, a new applicable categorization of Korean modality system viz. tense, aspect, mood, negation, and voice, will be proposed through a contrastive analysis of Chinese and Korean from the viewpoint of a practical MT system. In order to precisely determine the modal expression, effective feature selection frameworks for Chinese are presented with a variety of machine learning methods. As a result, our proposed approach achieved an accuracy of 83.10%. [ABSTRACT FROM AUTHOR]

The detection of a gap in relative clauses is essential in syntactic and semantic analysis of natural language processing. However, it is difficult to recognize whether a relative clause has a gap or it is gapless. Previous work related to relative clauses has been focusing mostly on theoretic linguistics without practically automatic classification, or classifying relative clauses using deep-level knowledge only available for a specific language. So, this paper proposes automatic classification method of a relative clause — whether it has a gap or gapless —, using easily obtainable features from any language. Features are extracted from the lexical forms and POS-tags in a relative clause, its headnoun, and contexts around a relative clause. Based on Support Vector Machines learning algorithm, our proposed method outperformed the baseline system by 25.11 percent. We also analyze the contribution rate of each simple feature to the classification, and the effect of contexts around a relative clause on the classification performance. [ABSTRACT FROM AUTHOR]

Word-spacing errors, one of the most frequent errors in Korean, produce ambiguities in the lexical interpretation of parts of speech or render sentences including them incomprehensible. Resolving those errors is thus crucial in Korean language processing application domains. In this paper, we propose a stochastic Korean word-spacing system with smoothing using Korean Spelling Checker, which is equally robust for both inner data and external data. In order to cope with various problems of word-spacing, this study (a) presents a simple stochastic word-spacing system with only two parameters using the odds favoring the inner-spacing of a given syllable bigram as well as relative word ftequencies, and (b) endeavors to (i) remove noise from the training data and (ii) diminish training data-dependency by dynamically creating a candidate word with a longest-radix-selecting algorithm. The system thus becomes robust against unseen words and offers a similar performance for both inner data and external data: it obtained a 98.47% and a 97.78% precision in word-unit correction for the inner test data and the balanced external test data, respectively. [ABSTRACT FROM AUTHOR]

Noun phrases are commonly used for generating index terms for information retrieval systems. Therefore, -we need an effective noun phrase extraction method. In this paper, we propose an approach to extract maximal noun phrases from Chinese text. Although previous studies have been proposed to extract noun phrases, most of them are only applicable to Western languages. To the best of our knowledge, very few has handled Chinese text. Many existing approaches for Western languages made use of statistical methods. However, due to the complicated structure of maximal Chinese noun phrase, pure statistical approaches are not effective. We attempt to improve the performance of a statistical method by integrating it with the transformation-based error-driven learning (TEL) technique. Our methodology includes two modules. The first module applies a statistical method to extract Chinese noun phrases. The performance of this approach, in terms of precision and recall, is investigated. The second module applies the TEL algorithm to further refine the output of the first module. The TEL algorithm automatically learns a set of transformation rules to fix the errors that are obtained through comparing the output of the first module with the correctly annotated corpus. The learned rules can be applied to sentences in any corpus one by one to correct the errors. The TEL algorithm is shown to be effective in improving the precision. [ABSTRACT FROM AUTHOR]

In natural language processing systems, it is very important to understand the meaning of time expressions and convert them into formal representations which include quantitative values. This paper proposes a method for understanding time expressions. For components of which time expressions consist, 31 concepts are introduced. Formal representations for these concepts are defined and an algorithm of analyzing the meaning of time expressions is proposed. The representation includes both conceptual and quantitative aspects in comparison with traditional approaches such as those of machine translation systems, and allows for more precise understanding of time expressions. An algorithm for translating time expressions into the corresponding formal representation by using a left-to-right parsing technique is presented. Experiments for about 2,000 time expressions extracted from actual documents show that the proposed method yields correct interpretation in about 93% of the cases. [ABSTRACT FROM AUTHOR]

The work presented in this paper is motivated by the practical need for content extraction, and the available data source and evaluation benchmark from the ACE program. The Chinese entity detection and tracking task is of particular interest to us. A novel solution is proposed to alleviate the language-independent and language-dependent problems special in this task. Mention detection takes advantages of machine learning approaches and character-based models. It manipulates different types of entities being mentioned and different constitution units (i.e., extents and heads) separately. Mentions referring to the same entity are linked together by integrating most-specific-first and closest-first rule based pairwise clustering algorithms. Types of mentions and entities are determined by head-driven classification approaches. The implemented system achieves 66.1 of ACE value, which has been one of the top-tier results. [ABSTRACT FROM AUTHOR]

Deep parsing of Chinese sentences is a very challenging task due to their complexity such as ambiguous word boundaries and meanings. An alternative mode of Chinese language processing is to perform shallow parsing of Chinese sentences in which chunk segmentation plays an important role. In this paper, we present a chunk segmentation algorithm using a combined statistical and rule-based approach (CSRA). The decision rules for refining chunk segmentation are generated from incorrectly segmented chunks from a statistical model which is built on a training corpus. Experimental results show that the CSRA works well and produces satisfactory chunk segmentation results for subsequent processes such as chunk tagging and chunk collocation extraction. [ABSTRACT FROM AUTHOR]

In this paper, we will describe a Korean transliterated foreign word extraction algorithm. In the proposed method, we reformulate the foreign word extraction problem as a syllable-tagging problem such that each syllable is tagged with a foreign syllable tag or a pure Korean syllable tag. Syllable sequences of Korean strings are modelled by Hidden Markov Model whose state represents a character with binary marking to indicate whether the syllable is part of a transliterated foreign word or not. The proposed method extracts a transliterated foreign word with high recall rate and precision rate. Moreover, our method shows good performance even with small-sized training corpora. [ABSTRACT FROM AUTHOR]

Document image compression is important in modern communication systems, and many lossless/lossy compression algorithms have been proposed for a variety of documents. In documents with Chinese characters, there are 5401 commonly used Chinese character patterns. It requires a large amount of space to store all Chinese character patterns of different fonts in computer systems. However, the compression of Chinese character patterns has not been extensively studied. In this paper, a new document image compression method is proposed. The purpose is to provide an effective encoding algorithm for Chinese character patterns and in the meantime obtain good compression results for general documents. The proposed method includes rectangular region partitioning, encoding of rectangular regions, and encoding of contour information. An input image is first partitioned into nonoverlapping blocks, and each block that contains black pixels is partitioned into nonoverlapping rectangles. The rectangles are then encoded in an effective fashion. For the purpose of lossless compression, contour information is exploited to encode the contour blocks with static Huffman coding. Experimental results showed that the proposed method is not only suitable for Chinese documents but also has good performance for general documents. [ABSTRACT FROM AUTHOR]

An extendible hashing scheme resolves bucket overflows by reorganizing the hash function and file structure locally, so it is very suitable for fast key retrieval of dynamic key sets. However, it cannot search keys that contain a given string as substrings efficiently. In this paper, in order to design this substring search in extendible hashing, signature vectors are introduced as hash values, and a trie structure as an extendible hash table, where each vector is composed by a bit stream. Pseudo signature vectors are defined to identify the buckets, and a constrained depth-first search is presented to traverse the arcs of the trie structure. To construct a compact trie despite an increase in the number of keys, uniform signature vectors are introduced, and the method for an incremental expansion of the hash table is proposed. This approach can restrict the size of the bit stream for each key, making constrained depth-first search efficient. From simulation results, by applying the presented schemes to Japanese key sets, it showed a decrease in the number of accessed buckets of 85% in comparison with traditional extendible hashing for which only descriptors were used. [ABSTRACT FROM AUTHOR]

Machine transliteration is automatic generation of the phonetic equivalents in a target language given a source language term, which is useful in many cross language applications. Transliteration between far distant languages, e.g. English and Chinese, is challenging because their phonological dissimilarities are significant. Existing techniques are typically rule-based or statistically noisy channel-based. Their accuracies are very low due to their intrinsic limitations on modeling transcription details. We propose direct statistical approaches on transliterating phoneme sequences for English–Chinese name translation. Aiming to improve performance, we propose two direct models: First, we adopt Finite State Automata on a process of direct mapping from English phonemes to a set of rudimentary Chinese phonetic symbols plus mapping units dynamically discovered from training. An effective algorithm for aligning phoneme chunks is proposed. Second, contextual features of each phoneme are taken into consideration by means of Maximum Entropy formalism, and the model is further refined with the precise alignment scheme using phoneme chunks. We compare our approaches with the noisy channel baseline that applies IBM SMT model, and demonstrate their superiority. [ABSTRACT FROM AUTHOR]