This paper describes our project on Japanese compound verbs.Japanese "Verb (adnominal form) + Verb" compounds, which are treated as single verbs, frequently appear in daily communication.They are not sufficiently registered in Japanese dictionaries or thesauri.We are now compiling a list of the synonymous expressions of compound verbs in "compound verb lexicon" built by the National Institute of Japanese Language and Linguistics.We extracted synonymous words and phrases of compound verbs from five hundred million Japanese web corpora.As a result, synonymous expressions of 1800 compound verbs were obtained automatically among 2700 in the "compound verb lexicon".From our data, we observed that some compound verbs represent not only motion but also additional nuances such as an emotional one.In order to reflect the abundant meanings that compound verbs own, we will try to think of a link of synonymous expressions to Japanese wordnet.Concretely, in the case of synonymous phrases, we try to link adverbial expressions which are a part of phrases to the adverbial synset in Japanese wordnet.

We are trying to construct a conceptual system that accurately represents human thoughts by fusing of semantic networks.As semantic networks to fuse, we use the Japanese Wordnet which is a thesaurus made manually based on linguistic intuition and the knowledge acquired automatically from the actual text stored in the huge corpus.Such knowledge are represented as mutual relations of the concepts of words.In order to acquire such relations, we focus on the case relations in sentences and calculate inclusive relations of co-occurrence by using Complementary Similarity Measure.As an application and verification of the conceptual system created, we try to simulate human associations by using the conceptual system.As an experimental result, we found the obvious difference in generated association links between using the semantic network of Japanese Wordnet and using the fused semantic networks with Japanese Wordnet and the acquired mutual relations.

Dependency parsing has become increasingly popular for a surge of interest lately for applications such as machine translation and question answering. Currently, several supervised learning methods can be used for training high-performance dependency parsers if sufficient labeled data are available. However, currently used statistical dependency parsers provide poor results for words separated by long distances. In order to solve this problem, this article presents an effective dependency parsing approach of incorporating short dependency information from unlabeled data. The unlabeled data is automatically parsed by using a deterministic dependency parser, which exhibits a relatively high performance for short dependencies between words. We then train another parser that uses the information on short dependency relations extracted from the output of the first parser. The proposed approach achieves an unlabeled attachment score of 86.52%, an absolute 1.24% improvement over the baseline system on the Chinese Treebank data set. The results indicate that the proposed approach improves the parsing performance for longer distance words.

Patent classification is a large-scale, hierarchical, imbalanced, multi-label problem. The number of samples in a real-world patent classification typically exceeds one million, and this number increases every year. An effective patent classifier must be able to deal with this situation. This paper discusses the use of min-max modular support vector machine (M3-SVM) to deal with large-scale patent classification problems. The method includes three steps: decomposing a large-scale and imbalanced patent classification problem into a group of relatively smaller and more balanced two-class subproblems which are independent of each other, learning these subproblems using support vector machines (SVMs) in parallel, and combining all of the trained SVMs according to the minimization and the maximization rules. M3-SVM has two attractive features which are urgently needed to deal with large-scale patent classification problems. First, it can be realized in a massively parallel form. Second, it can be built up incrementally. Results from experiments using the NTCIR-5 patent data set, which contains more than two million patents, have confirmed these two attractive features, and demonstrate that M3-SVM outperforms conventional SVMs in terms of both training time and generalization performance.

We describe a novel approach for validating transliteration hypotheses based on a Web mining technique. We implemented a machine transliteration system and generated Chinese, Japanese, and Korean transliteration hypotheses for given English words. Then, we mined the Web for features relevant to validating transliteration hypotheses. Finally we validated transliteration hypotheses using machine learning algorithms learned with the mined features. Comparing Web counts with our Web mining technique, our proposed method consistently performed better than systems based on Web counts, regardless of the language.

Japanese dependency structure is usually represented by relationships between phrasal units called bunsetsus. One of the biggest problems with dependency structure analysis in spontaneous speech is that clause boundaries are ambiguous. This paper describes a method for detecting the boundaries of quotations and inserted clauses and that for improving the dependency accuracy by applying the detected boundaries to dependency structure analysis. The quotations and inserted clauses are determined by using an SVM-based text chunking method that considers information on morphemes, pauses, fillers, etc. The information on automatically analyzed dependency structure is also used to detect the beginning of the clauses. Our evaluation experiment using Corpus of Spontaneous Japanese (CSJ) showed that the automatically estimated boundaries of quotations and inserted clauses helped to improve the accuracy of dependency structure analysis.

The Web provides the largest data collection, which reflects language use in daily life. With the advent of new technology and the flood of information on the Web, it has become quite common to create new terms supporting new concepts and translate these terms into non-Latin languages with "transliteration" referring to "translation by sound". Cross-language natural language processing applications, such as machine translation and cross-language information retrieval, usually need a translation dictionary, which affects the quality of the applications. However; the transliteration lexicons are usually unregistered in the translation dictionary. To address the problem, we present a transliteration lexicon acquisition model that mines the Web for transliteration lexicons. In this paper, we describe techniques of comparing phonetic-similarity to recognize transliteration pair candidates on the Web and of finding the correct transliteration pairs based on joint-validation. The techniques were evaluated against manually constructed transliteration lexicons. Our experiments revealed that the techniques effectively found transliteration lexicons on the Web

Machine transliteration is a method for automatically converting words in one language into phonetically equivalent ones in another language. Machine transliteration plays an important role in natural language applications such as information retrieval and machine translation, especially for handling proper nouns and technical terms. Four machine transliteration models -- grapheme-based transliteration model, phoneme-based transliteration model, hybrid transliteration model, and correspondence-based transliteration model -- have been proposed by several researchers. To date, however, there has been little research on a framework in which multiple transliteration models can operate simultaneously. Furthermore, there has been no comparison of the four models within the same framework and using the same data. We addressed these problems by 1) modeling the four models within the same framework, 2) comparing them under the same conditions, and 3) developing a way to improve machine transliteration through this comparison. Our comparison showed that the hybrid and correspondence-based models were the most effective and that the four models can be used in a complementary manner to improve machine transliteration performance.

Machine transliteration is an automatic method for converting words in one language into phonetically equivalent ones in another language. There has been growing interest in the use of machine transliteration to assist machine translation and information retrieval. Three types of machine transliteration models---grapheme-based, phoneme-based, and hybrid---have been proposed. Surprisingly, there have been few reports of efforts to utilize the correspondence between source graphemes and source phonemes, although this correspondence plays an important role in machine transliteration. Furthermore, little work has been reported on ways to dynamically handle source graphemes and phonemes. In this paper, we propose a transliteration model that dynamically uses both graphemes and phonemes, particularly the correspondence between them. With this model, we have achieved better performance---improvements of about 15 to 41% in English-to-Korean transliteration and about 16 to 44% in English-to-Japanese transliteration---than has been reported for other models.

We investigated, via experiment, knowledge of normative honorific expressions as used in textbooks and in practice by people. Forty subjects divided into four groups according to age (younger/older) and gender (male/female) participated in the experiments. The results show that knowledge about the use of normative honorific expressions in textbooks is similar to that demonstrated by the younger subject groups, but differed from that of the older subject groups. The knowledge of the older subjects was more complex than that shown in textbooks or demonstrated by the younger subjects. A model that can identify misuse of honorific expressions in sentences is the framework for this investigation. The model is minimal, but could represent 76% to 92% of the subjects' knowledge regarding each honorific element. This model will be useful in the development of computer-aided systems to help teach how honorific expressions should be used.

We present a series of experiments designed to compare grammar-based and robust approaches to speech understanding, performed in the context of an Open Source medical speech translation system. We used two versions of the system, one grammar-based and one robust, trained off the same training data, and evaluated them on test data collected using both versions of the system. The experiments were constructed so as to avoid several methodological problems which occurred in earlier work reported in the literature. We found that the grammarbased version gave significantly better results than the robust version, with the difference increasing as subjects became more familiar with the system's coverage. The rate of improvement in subject performance was positively affected by providing them with an intelligent online help system.

This paper claims that constructing a dictionary using bilingual pairs obtained from parallel corpora needs not only correct alignment of two noun phrases but also judgment of its appropriateness as an entry. It specifically addresses the latter task, which has been paid little attention. It demonstrates a method of selecting a suitable entry using Support Vector Machines, and proposes to regard as the features the common and the different parts between a current translation and a new translation. Using experiment results, this paper examines how selection performances are affected by the four ways of representing the common and the different parts: morphemes, parts of speech, semantic markers, and upper-level semantic markers. Moreover, we used n-grams of the common and the different parts of above four kinds of features. Experimental result found that representation by morphemes marked the best performance, F-measure of 0.803.

In this paper, we present evidence that providing users of a speech to speech translation system for emergency diagnosis (MedSLT) with a tool that helps them to learn the coverage greatly improves their success in using the system. In MedSLT, the system uses a grammar-based recogniser that provides more predictable results to the translation component. The help module aims at addressing the lack of robustness inherent in this type of approach. It takes as input the result of a robust statistical recogniser that performs better for out-of-coverage data and produces a list of in-coverage example sentences. These examples are selected from a defined list using a heuristic that prioritises sentences maximising the number of N-grams shared with those extracted from the recognition result.

We describe a method for automatically rating the machine translatability of a sentence for various machine translation (MT) systems. The method requires that the MT system can bidirectionally translate sentences in both source and target languages. However, it does not require reference translations, as is usual for automatic MT evaluation. By applying this method to every component of a sentence in a given source language, we can automatically identify the machine-translatable and non-machinetranslatable parts of a sentence for a particular MT system. We show that the parts of a sentence that are automatically identified as nonmachine-translatable provide useful information for paraphrasing or revising the sentence in the source language, thus improving the quality of the final translation.

Automatic word alignment is an important technology for extracting translation knowledge from parallel corpora. However, automatic techniques cannot resolve this problem completely because of variances in translations. We therefore need to investigate the performance potential of automatic word alignment and then decide how to suitably apply it. In this paper we first propose a lexical knowledge-based approach to word alignment on a Japanese-Chinese corpus. Then we evaluate the performance of the proposed approach on the corpus. At the same time we also apply a statistics-based approach, the well-known toolkit GIZA++, to the same test data. Through comparison of the performances of the two approaches, we propose a multi-aligner, exploiting the lexical knowledge-based aligner and the statistics-based aligner at the same time. Quantitative results confirmed the effectiveness of the multi-aligner.

We are constricting a Japanese-Chinese parallel corpus, which is a part of the NICT Multilingual Corpora. The corpus is general domain, of large scale of about 40,000 sentence pairs, long sentences, annotated with detailed information and high quality. To the best of our knowledge, this will be the first annotated Japanese-Chinese parallel corpus in the world. We created the corpus by selecting Japanese sentences from Mainichi Newspaper and then manually translating them into Chinese. We then annotated the corpus with morphological and syntactic structures and alignments at word and phrase levels. This paper describes the specification in human translation and detailed information annotation, and the tools we developed in the project. The experience we obtained and points we paid special attentions are also introduced for share with other researches in corpora construction.

In recent years, various types of tagged corpora have been constructed and much research using tagged corpora has been done. However, tagged corpora contain errors, which impedes the progress of research. Therefore, the correction of errors in corpora is an important research issue. In this study we investigate the correction of such errors, which we call corpus correction. Using machine-learning methods, we applied corpus correction to a verb modality corpus for machine translation. We used the maximum-entropy and decision-list methods as machine-learning methods. We compared several kinds of methods for corpus correction in our experiments, and determined which is most effective by using a statistical test. We obtained several noteworthy findings: (1) Precision was almost the same for both detection and correction, so it is more convenient to do both correction and detection, rather than detection only. (2) In general, the maximum-entropy method worked better than the decision-list method; but the two methods had almost the same precision for the top 50 pieces of extracted data when closed data was used. (3) In terms of precision, the use of closed data was better than the use of open data; however, in terms of the total number of extracted errors, the use of open data was better than the use of closed data. Based on our analysis of these results, we developed a good method for corpus correction. We confirmed the effectiveness of our method by carrying out experiments on machine translation. As corpus-based machine translation continues to be developed, the corpus correction we discuss in this article should prove to be increasingly significant.

This work presents a part-versus-part decomposition method for massively parallel training of multi-class support vector machines (SVMs). By using this method, a massive multi-class classification problem is decomposed into a number of two-class subproblems as small as needed. An important advantage of the part-versus-part method over existing popular pair wise-classification approach is that a large-scale two-class subproblem can be further divided into a number of relatively smaller and balanced two-class subproblems, and fast training of SVMs on massive multi-class classification problems can be easily implemented in a massively parallel way. To demonstrate the effectiveness of the proposed method, we perform simulations on a large-scale text categorization problem. The experimental results show that the proposed method is faster than the existing pairwise-classification approach, better generalization performance can be achieved, and the method scales up to massive, complex multi-class classification problems.

This paper describes two methods for detecting word segments and their morphological information in a Japanese spontaneous speech corpus, and describes how to tag a large spontaneous speech corpus accurately by using the two methods. The first method is used to detect any type of word segments. The second method is used when there are several definitions for word segments and their POS categories, and when one type of word segments includes another type of word segments. In this paper, we show that by using semi-automatic analysis, we achieve a precision of better than 99% for detecting and tagging short-unit words and 97% for long-unit words; the two types of words that comprise the corpus. We also show that better accuracy is achieved by using both methods than by using only the first.

This paper describes two methods for detecting word segments and their morphological information in a Japanese spontaneous speech corpus, and describes how to tag a large spontaneous speech corpus accurately by using the two methods. The first method is used to detect any type of word segments. The second method is used when there are several definitions for word segments and their POS categories, and when one type of word segments includes another type of word segments. In this paper, we show that by using semi-automatic analysis we achieve a precision of better than 99% for detecting and tagging short words and 97% for long words; the two types of words that comprise the corpus. We also show that better accuracy is achieved by using both methods than by using only the first.

We have aligned Japanese and English news articles and sentences to make a large parallel corpus. We first used a method based on cross-language information retrieval (CLIR) to align the Japanese and English articles and then used a method based on dynamic programming (DP) matching to align the Japanese and English sentences in these articles. However, the results included many incorrect alignments. To remove these, we propose two measures (scores) that evaluate the validity of alignments. The measure for article alignment uses similarities in sentences aligned by DP matching and that for sentence alignment uses similarities in articles aligned by CLIR. They enhance each other to improve the accuracy of alignment. Using these measures, we have successfully constructed a large-scale article and sentence alignment corpus available to the public.

The elastic-input neuro-tagger and hybrid tagger, combined with a neural network and Brill's error-driven learning, have already been proposed to construct a practical tagger using as little training data as possible. When a small Thai corpus is used for training, these taggers have tagging accuracies of, respectively, 94.4% and 95.5% (accounting only for the ambiguous words that relate to the parts of speech). In this study, in order to construct more accurate taggers, we developed new tagging methods using three different machine-learning approaches: the decision list, maximum entropy, and the support vector machine methods. We then performed tagging experiments using them. Our results show that the support vector machine method has the best precision (96.1%), and that it is capable of improving the accuracy of tagging in the Thai language. The improvement in accuracy was also confirmed by using a statistical test (a sign test). Finally, we examined theoretically all these methods in an effort to determine how the improvements were achieved. We found that the improvements were due to our use of word information, which is helpful for tagging, and a support vector machine that performed well.

We propose a statistical method that finds the maximum-probability segmentation of a given text. This method does not require training data because it estimates probabilities from the given text. Therefore, it can be applied to any text in any domain. An experiment showed that the method is more accurate than or at least as accurate as a state-of-the-art text segmentation system.

This paper describes an evaluation experiment designed to determine groups of subjects who prefer reading MT outputs to reading the original text. Our approach can be applied to any language pairs, but we will explain the methodology by taking English to Japanese translation as an example. In the case of E-J MT, it can be assumed that main users are Japanese and that most of them have some knowledge of English. It is often the case, in the case of E-J MT systems, that those people who are comfortable with reading English do not find E-J MT outputs useful, and in many cases, they would rather prefer reading the original English text. On the other hand, E- J MT outputs prove to be useful to those who find it hard to read the original English texts. We have used the reading comprehension part of the Test Of English for International Communication (TOEIC) to determine the threshold English ability level, dividing these two user groups.

% " & ' ( ) * + -, ./ ( * )0 1 2 * 3 4 2 *$ 65 7 8 ) & :9 ;, <0 $ = ?>A@ CB D FE G %E IH KJ ML N> PO IB D QB SR 2T U> PV W YX E ZR

As a beginning study on self-organizing a general Japanese semantic map, which will be very useful in natural language processing, particularly in document organization and information retrieval, this paper describes the construction of a semantic map of Japanese nouns mapped according to their adnominal constituents. These maps are not only an important part of a general Japanese semantic map that we aim to construct, but can also be a powerful tool for supporting the analysis of the relation between head nouns and their adnominal constituents, an important issue in studies of Japanese pragmatics.

Although high recognition accuracy can be obtained for speech in the form of reading a written text or similar by using state-of-the art speech recognition technology, the accuracy is quite poor for freely spoken spontaneous speech. From this perspective, a five-year national project for raising the technological level of speech recognition and understanding commenced in Japan in 1999. The project focuses on building a large-scale spontaneous speech corpus and acoustic and linguistic modeling for spontaneous speech recognition and summarization. This paper reports some results of preliminary experiments which have been conducted at Tokyo Institute of Technology. Experimental results show that acoustic and language modeling based on the actual spontaneous speech corpus is far more effective than modeling based on read speech. It was also shown that our proposed automatic speech summarization method could effectively extract relatively important information and remove redundant and irrelevant information. 1.

We exploit and extend the Generative Lexicon Theory to develop a formal description of adnominal constituents in a lexicon which can deal with linguistic phenomena found in Japanese adnominal constituents. We classify the problematic behavior into "static disambiguation" and "dynamic disambiguation" tasks. Static disambiguation can be done using lexical information in a dictionary, whereas dynamic disambiguation requires inferences at the knowledge representation level.

This paper presents a part of speech (POS) neuro tagger which consists of a 3-layer perceptron with elastic input. Computer experiments show that the neuro tagger has an accuracy of 94.4% for tagging ambiguous words when a small Thai corpus with 22,311 ambiguous words is used for training. A series of comparative experiments further show that the neuro tagger is definitely far superior to the statistical models including the frequency model (a base-line model), local n-gram model, and HMM.

We are developing an Intelligent Network News Reader which extracts news articles for the user.In contrast to ordinary information retrieval and abstract generation, this method utilizes an "information context" to select articles from newsgroups on the intemet.The salient feature of this system is that it retrieves articles dynamically, adapting to the user's interests, not classifying them beforehand.Since this system measures the semantic distance between articles, it is possible to refer to the necessary information without being constrained within a particular news group.We will finish our prototype of the Intelligent Network News Reader in March 1997 and its practical version in