© 2020, Springer Nature Singapore Pte Ltd. A sentence encoder that can be readily employed in many applications or effectively fine-tuned to a specific task/domain is highly demanded. Such a sentence encoding technique would achieve a broader range of applications if it can deal with almost arbitrary word-sequences. This paper proposes a training regime for enabling encoders that can effectively deal with word-sequences of various kinds, including complete sentences, as well as incomplete sentences and phrases. The proposed training regime can be distinguished from existing methods in that it first extracts word-sequences of an arbitrary length from an unlabeled corpus of ordered or unordered sentences. An encoding model is then trained to predict the adjacency between these word-sequences. Herein an unordered sentence indicates an individual sentence without neighboring contextual sentences. In some NLP tasks, such as sentence classification, the semantic contents of an isolated sentence have to be properly encoded. Further, by employing rather unconstrained word-sequences extracted from a large corpus, without heavily relying on complete sentences, it is expected that linguistic expressions of various kinds are employed in the training. This property contributes to enhancing the applicability of the resulting word-sequence/sentence encoders. The experimental results obtained from supervised evaluation tasks demonstrated that the trained encoder achieved performance comparable to existing encoders while exhibiting superior performance in unsupervised evaluation tasks that involve incomplete sentences and phrases.

<p>This paper describes a spoken dialogue system for accommodating a user's information behaviors with various levels of information need. Our system, given a set of same-topic news articles, compiles a utterance plan that consists of a primary plan for delivering main news content, and the associated subsidiary plans for supplementing the main content. A primary plan is generated by applying text summarization and style conversion techniques. The subsidiary plans are compiled by considering potential user/system interactions. To make this mechanism work, we first classified user's possible passive/active behaviors, and then designed the corresponding system actions. We empirically confirmed that our system was able to deliver the news content smoothly while dynamically adapting to the change of user's intention levels. The smoothness of a conversation can be attributed to the pre-compiled utterance plan.</p>

© 2018, Springer Nature Switzerland AG. The performances of an ad-hoc video search (AVS) task can only be improved when the video processing for analyzing video contents and the linguistic processing for interpreting natural language queries are nicely combined. Among the several issues associated with this challenging task, this paper particularly focuses on the sense disambiguation/filtering (WSD/WSF) of the terms contained in a search query. We propose WSD/WSF methods which employ distributed sense representations, and discuss their efficacy in improving the performance of an AVS system which makes full use of a large bank of visual concept classifiers. The application of a WSD/WSF method is crucial, as each visual concept classifier is linked with the lexical concept denoted by a word sense. The results are generally promising, outperforming not only a baseline query processing method that only considers the polysemy of a query term but also a strong WSD baseline method.

© 2016 ACM. Passive consumption of information is boring in most cases and even painful in some cases, especially when the information content is delivered by employing speech media. The user of a speech-based information delivery system, for example a text-to-speech system, usually cannot interrupt the ongoing information ow, inhibiting her/him to confirm some part of the content, or to pose an inquiry for further information exploration. We argue that a carefully designed spoken dialog system could remedy these undesirable situations, and further enable an enjoyable conversation with the users. The key technologies to realize such an attractive dialog system are: (1) pre-compilation of a dialog plan based on the analysis of a source content, and (2) the dynamic recognition of user's state of understanding and interests. This paper illustrates technical views to implement these functionalities, and discusses a dialog example to exemplify the technical merits of the proposed system.

This paper describes our independent effort for extending the monolingual semantic textual similarity (STS) task setting to multiple cross-lingual settings involving English, Japanese, and Chinese. So far, we have adopted a "monolingual similarity after translation" strategy to predict the semantic similarity between a pair of sentences in different languages. With this strategy, a monolingual similarity method is applied after having (one of) the target sentences translated into a pivot language. Therefore, this paper specifically details the required and developed resources to implement this framework, while presenting our current results for English-Japanese-Chinese cross-lingual STS tasks that may exemplify the validity of the framework.

Given lexical-semantic resources in different languages, it is useful to establish cross-lingual correspondences, preferably with semantic relation labels, between the concept nodes in these resources. This paper presents a framework for enabling a cross-lingual/node-wise alignment of lexical-semantic resources, where cross-lingual correspondence candidates are first discovered and ranked, and then classified by a succeeding module. Indeed, we propose that a two-tier classifier configuration is feasible for the second module: the first classifier filters out possibly irrelevant correspondence candidates and the second classifier assigns a relatively fine-grained semantic relation label to each of the surviving candidates. The results of Japanese-to-English alignment experiments using EDR Electronic Dictionary and Princeton WordNet are described to exemplify the validity of the proposal.

© 1963-2018 ACL. Evocation is a directed yet weighted semantic relationship between lexicalized concepts. Although evocation relations are considered potentially useful in several semantic NLP tasks, the prediction of the evocation relation between an arbitrary pair of concepts remains difficult, since evocation relationships cover a broader range of semantic relations rooted in human perception and experience. This paper presents a supervised learning approach to predict the strength (by regression) and to determine the directionality (by classification) of the evocation relation that might hold between a pair of lexicalized concepts. Empirical results that were obtained by investigating useful features are shown, indicating that a combination of the proposed features largely outperformed individual baselines, and also suggesting that semantic relational vectors computed from existing semantic vectors for lexicalized concepts were indeed effective for both the prediction of strength and the determination of directionality.

This paper demonstrates that Wordnet-LMF, a version of ISO LMF, allows us to effectively design and implement Web services for accessing WordNet-type semantic lexicons that conform to the REST Web service architecture. The implemented prototype service currently provides access to native wordnets as well as to a bilingual concept dictionary. This paper thus describes slight revisions that were made to the Wordnet-LMF specifications to model and accommodate a non-wordnet-native bilingual concept dictionary. © 2012 Springer Science+Business Media B.V.

© 2011, Springer-Verlag Berlin Heidelberg. The Language Grid is a distinctive language service infrastructure in the sense that it accommodates a wide variety of user needs, ranging from technical novices to experts; language resource consumers to language resource providers. As these language services are various in type and each of them can be idiosyncratic in many aspects, the service infrastructure has to address the issue of interoperability. A key to solve this issue is not only to build the services around standardized resources and interfaces, but also to establish a knowledge structure that copes effectively with a range of language services. Given this knowledge structure, referred to as a service ontology, each language service can be systematically classified and its usage specified by a corresponding API. This not only enables the utilization of existing language resources but facilitates the dissemination of newly created language resources as services.

This paper proposes a framework for representing cross-lingual/interlingual lexical semantic correspondences that are expected to be recovered through a series of on-demand/on-the-fly invocations of a lexical semantic matching process. One of the central notions of the proposed framework is a pseudo synset, which is introduced to represent a cross-lingual/multilingual lexical concept, jointly denoted by word senses in more than one language. Another important ingredient of the proposed framework is a framework for semantifying bilingual lexical resource entries. This is a necessary substep when associating and representing corresponding lexical concepts in different languages by using bilingual lexical resources. Based on these devices, this paper further discusses possible extensions to the ISO standard lexical markup framework (LMF). These extensions would enable recovered correspondences to be organized as a dynamic secondary language resource, while keeping the existing primary language resources intact.

異文化コラボレーションの様々な局面におけるコミュニケーションを支援するための有用な要素としての言語資源の重要性に対する認識が高まっており,既存の汎用的な言語資源や,新たに特定の目的に即して構築される言語資源を適切に組み合わせて利用できる基盤の実現が強く求められている.オープンなネットワーク環境において,言語資源が適切に利用され得るためには,再利用性と相互運用性がキーとなる.言語資源に対して共通的な語彙によるメタ記述が与えられていること,また,言語資源の内容がそのタイプに応じてモデル化され,それに応じた標準的なアクセス手段で利用可能であることが重要である.本サーベイ論文は,このような再利用性と相互運用性を達成するための言語資源の記述とモデル化の枠組みを整理して提示する.まず,言語資源の全体に関するメタ記述の枠組みを示す.次に,言語資源の内容に関する相互運用性に関連し,言語的注釈の記述とモデル化のための枠組みについて述べる.これらの枠組みが機能するためには,言語資源の作成者,利用者などの関連者による枠組みの共通理解が必須であり標準化が重要となる.このため,関連した国際標準化の動向についても述べる.

本稿では,セマンティックWebと言語資源(LR:language resource)・言語技術(LT:language technology)の互恵的な関係について述べる.セマンティックWebにおいては,利用者が共通に理解できる「意味」をWeb上のさまざまなリソースにメタデータとして与えることが必要となる.言語技術の進展により,リソースから意味的な情報を抽出することが可能となってきており,抽出された意味情報を対象世界において標準化されたオントロジーと対応付けることにより,利用者にとって共通に理解できる(意味的相互運用性のある)メタデータを生成することができる.一方で,セマンティックWebは言語資源や言語技術の適用領域を飛躍的に拡大する可能性がある。いわゆるWebサービスの関連技術により,辞書やコーパスなどの言語資源や,翻訳や解析などの言語処理機能をネットワーク上で組み合わせて利用することが可能となっている.さらに,これらの構成要素に意味的相互運用性のある記述を与えておくことにより,ユーザの要求に応じて、要素の有意な組合せによって所望の機能を提供するWebサービスを効率よく実現することが可能となる.本稿では,セマンティックWebと言語資源・言語技術に関する上記の2つの方向性について概観する.前者に関しては、特に,言語的なオントロジーといわゆる上位オントロジーの対応付け,各国語の言語的なオントロジーの連携について述べる.後者に関しては,言語資源,言語処理機能,さらには,これらの対象となる言語的データ・情報を記述するための標準的な枠組みの必要性について述べる.

© 2018 IEEE. To ensure that humans and robots can safely coexist, the ability to recognize human behavior is a prerequisite for robots and a fundamental technical challenge for researchers. Current research can only recognize relatively simple cases of human behavior due to the lack of enough data and archetypally designed experiments. Our study elucidates human behavior in a systematic manner by observing the behavior of human subjects under more complex situations where they are surrounded by other people or objects to address the challenge. We focus on the following common situation that people pass each other through a narrow passage. We constructed a motion capture room with a narrow passage environment and measured the motion of human subjects performing different tasks. In addition, during the narrow passage experiment, we made subjects hold different daily necessaries (such as a backpack) to observe influences on human behavior. Our study found that passing and avoidance behavior exhibited by each of our subjects were significantly influenced by what kind of daily necessaries subjects carry. This research provides novel findings on human behavior in complex environments: in the case where subjects holding a handbag (a type of daily necessaries that stays next to one's body), they showed the tendency to be affected by the other subject and move more dynamically compared to the subject without anything or with other daily necessaries; in the case where subjects carrying a backpack (a type of daily necessaries on one's back) and looking at a smartphone, they also showed the tendency of being affected by others, but their motion is restricted compared to the subject without anything.

Copyright © TRECVID 2018.All rights reserved. The Waseda Meisei team participated in the TRECVID 2018 Ad-hoc Video Search (AVS) task [1]. For this year's AVS task, we submitted both manually assisted and fully automatic runs. Our approach focuses on the concept-based video retrieval, based on the same approach as last year. Specifically, it improves on the word-based keyword extraction method presented in last year's system, which could neither handle keywords related to motion nor appropriately capture the meaning of phrases or whole sentences in queries. To deal with these problems, we introduce two new measures: (i) calculating the similarity between the definition of a word and an entire query sentence, (ii) handling of multi-word phrases. Our best manually assisted run achieved a mean average precision (mAP) of 10.6%, which was ranked the highest among all submitted manually assisted runs. Our best fully automatic run achieved an mAP of 6.0%, which ranked sixth among all participants.

The recent research direction toward multimodal semantic representation would be further advanced, if we could have a machinery to collect adequate images from the Web, given a target concept. With this motivation, this paper particularly investigates into the Webimageabilities of the behavioral features (e.g. "beaver builds dams") of a basic-level concept (beaver). The term Web-imageability denotes how adequately the images acquired from the Web deliver the intended meaning of a complex concept. The primary contributions made in this paper are twofold: (1) "beaver building dams"-type queries can better yield relevant Web images, suggesting that the present participle form ("ing" form) of a verb ("building"), as a query component, is more effective than the base form; (2) the behaviors taken by animate beings are likely to be more depicted on the Web, particularly if the behaviors are, in a sense, inherent to animate beings (e.g., motion, consumption), while the creation-type behaviors of inanimate beings are not. The paper further analyzes linguistic annotations that were independently given to some of the images, and discusses an aspect of the semantic gap between image and language.

Interactive search environments are gaining more attractions than ever, particularly in the setting of cross-language information retrieval, where the query language and the document language are different. This paper, through experiments, shows that images acquired from the Web toward a translation of a user-issued query term can serve as an effective clue in selecting proper translations; hence they may be effective in improving the cross-language search effectiveness. This paper further discusses that the semantic feature and the sense familiarity of a query term could affect the utility of the associated images obtained from the Web.

This paper proposes a method to compute cross-lingual semantic similarity between synonym sets. By making use of Princeton Annotated Gloss Corpus as the source of target language statistics, the proposed method exhibited promising results in the experiments: More than 73% of the Princeton WordNet synsets were successfully recovered within the top-5 candidates, given a corresponding set of Japanese WordNet synsets. As the proposed method minimally requires that the input to be seen as an apparently synonymous word set, the method could be extended and the performance would be further improved by incorporating richer information such as textual glosses and/or structural constraints posed by the lexical resources at hand. © Christiane Fellbaum, Piek Vossen, 2012.

It is often argued that a set of standard linguistic processing functionalities should be identified, with each of them given a formal specification. We would benefit from the formal specifications; for example, the semi-automated composition of a complex language processing workflow could be enabled in due time. This paper extracts a standard set of linguistic processing functionalities and tries to classify them formally. To do this, we first investigated prominent types of language Web services/linguistic processors by surveying a Web-based language service infrastructure and published NLP toolkits. We next induced a set of standard linguistic processing functionalities by carefully investigating each of the linguistic processor types. The standard linguistic processing functionalities was then characterized by the input/output data types, as well as the required data operation types, which were also derived from the investigation. As a result, we came up with an ontological depiction that classifies linguistic processors and linguistic processing functionalities with respect to the fundamental data operation types. We argue that such an ontological depiction can explicitly describe the functional aspects of a linguistic processing functionality.

Servicization of language resources in a Web-based environment has opened up the potential for dynamically combined virtual lexical resources. Evolving lexical linked data could be realized, provided being recovered/discovered links among lexical resources are properly organized and maintained. This position paper examines a scenario, in which lexical semantic resources are cross-linguistically enriched, and sketches how this scenario could come about while discussing necessary ingredients. The discussions naturally include how the existing lexicon modeling framework could be applied and should be extended.

会議音声から話題の階層構造を抽出することにより議事録を生成するシステムを提案する.本システムでの議事録とは,大項目の下に小項目が列挙されているような階層的な話題構成をとるもので,各項目は簡潔な要約文になっているものである.話題構造の抽出は,会議音声の認識結果を単語ベクトルの系列ととらえ,この系列を話題単位に分割し,得られた話題単位の集合を階層的にクラスタリングすることによって行う.話題構造抽出処理の評価実験により,提案システムの手法は音声認識結果に対し頑健であり,また認識信頼度の低い単語を除去することにより更に精度が向上することを確認した.

Translation refinement is often observed when users communicate via machine translation systems. In this study, we analyzed user's translation refinement process through a controlled experiment. In the experiment, users translated sentences using a Japanese-English-Japanese turn-back translation. From the analysis, we discovered the following results: 1) The more knowledge users had about the source language, the better users could refine the original text, 2) Rule instruction was very effective in user's adaptation. Users who were reminded of the rules refined the original text ahead of other users, 3) Instructing operational rules were effective in helping user's adapation, while conditional rules were not as much effective. 4) Rule instruction was most effective to those who had midium knowledge in their source languages.

This paper analyzes the impact of user adaptation in MT-mediated communication. It clarifies how the user adapts to machine translation and how effective the adaptation is in terms of communication when the purpose of communication is clear.The most common alterations and their effectiveness strongly depend on the translation language pairs.In the case of Japanese-to-English translation, we observed two main alterations: replacing words or phrases to offset the difference in concepts between Japanese and English and supplementing subjects to offset the difference in modes of expression between Japanese and English.Since Korean and Japanese are similar languages, Korean users exhibited similar adaptation tendencies.The adaptation performed by Japanese users when referring to the English translation was very effective in improving the quality of the English translations.However, it was not so effective for Chinese and even less effective for Korean translations.

With the spread of mobile phones that can access the Internet, most people are forwarding their incoming email to their mobile phones. Mobile phones enable heavy users of emails to access them anywhere. The restricted capacities (display size and message limits) of mobile phones, however, prohibit whole messages from being read as is. We propose a new rulebased method of sentence extraction which takes the features of emails into consideration and describe its application to an email summarizer for mobile phones. Our method uses the expression and style specific to emails. The sentences extracted by the proposed method are virtually the same as those selected manually by majority voting. A task-oriented evaluation of the proposed method reveals its high precision and recall in screening for emails through mobile phones given a practical message size limit.

© 2004 EUSIPCO. This paper proposes a new segmentation procedure to detect audio source intervals for automatic indexing of broadcast news. The procedure is composed of an audio source detection part and a part that smoothes the detected sequences. The detection part uses three new acoustic feature parameters that are based on spectral cross-correlation: spectral stability, white noise similarity, and sound spectral shape. These parameters make it possible to capture the audio sources more accurately than can be done with conventional parameters. The smoothing part has a new merging method that drops erroneous detection results of short duration. Audio source classification experiments are conducted on broadcast news segments. Performance is increased by 6.6% when the proposed parameters are used and by 3.1% when the proposed merging method is used, showing the usefulness of our approach. Experiments confirm the impact of this proposal on broadcast news indexing.

Calculating the similarity between sentences, documents, or sets of documents precisely is a very important technique for text retrieval, information filtering, document classification, document clustering and so on. We propose a new model to calculate this similarity; we call it Graph-based text representation model (Graph model). In the Graph model, each document is translated into a subject graph that treats the significance of each word as its node&#039;s weight and the significance of each term-term association as its link&#039;s weight. The similarity between documents is calculated by the similarity between these graphs. This paper describes the concept and the realization of the Graph model, and evaluates it in a text retrieval system. Through the evaluation, we show that the Graph model is more effective than the Vector space model in calculating the similarity between documents, especially when each document has many topics such as Web pages, or sets of documents such as user profiles, category profiles and clusters. In addition, the similarity calculation is fast enough for text retrieval systems, given the situation where the index of the association is pre-compiled.

技術文章においては,内容をわかりやすく,かつ正確に伝達することが要求される.本論文では,上記のような要求条件を満たす技術文章を作成するためのテクニカルライティングのノウハウの分析に基づき,表現の書換え支援機能をもつ日本文推敲支援システムの構成方式を提案し,その有効性を評価する.システムは,3レベルの処理からなる.まず最初のレベルでは,文の形態素解析結果に基づいて長文の分割を行う.次のレベルではスタイル上の問題点を検出し,これらの修正を部分的な構文構造の変換を繰り返すことにより行う.最後のレベルでは文の正確な理解を妨げる可能性のある読点付与と語順の不適切さを文全体の構文構造を変換することにより修正する.以上の構成をもつシステムがわかりやすく正確な理解のしやすい文章を作成するという目的に対してどの程度有効であるかを,前者については読みやすさの指標を,後者については計算機による係り受け解析において生じる多義数をその近似指標として用いて評価した.その結果,本システムを用いた推敲作業実験により,対象文章の読みやすさの向上と,各文の構文的なあいまいさの減少を定量的に確認することができ,本システムの基本的な有効性を確認することができた.

It is important for structural analysis of texts and texts understanding to analyze cohesive relations between adjacent sentences. This paper proposes a new method to analyze cohesive relations between adjacent Japanese sentences based on constraints naturally contained in objective texts. We especially focused on explanative texts such as technical manuals. In these texts, the world to be explained constrains the sentence type and sentence adjacent patterns. For example, text for explaining a software command will describe some user operations and the associated system responses in some cohesive ways. We have developed a sentence type system suitable for software manuals and constructed a set of analysis rules. An experimental result shows that correct rate of our algorithm is about 80%.

計算機による日本語文書の推敲支援において 長文の推敲機能を実現するためには 文を分割することによって適切な文に書き換える文分割の技術を確立する必要がある本論文では 複文を対象とした接続構造解析に基づく文分割方式を提案する本方式では 接続形式の導く従属句の独立性に着目することによって 述語間の接続構造を解析し 分割点となる述語を決定するこのとき 連用中止表現を形態的特徴によって分類し 接続構造の解析に利用するこれらによって 計算コストの大きい文全体の構文解析を行わずに分割すべき述語を特定することができるまた 連用中止表現を含む文に対しても 高い精度で接続構造を解析することができるさらに 分割文を生成する岸には 連用中止表現と文末述語の形態的な特徴に着目することによって 連用中止表現で接続された単文間の意味的な関係を判定し その結果に応じた分割文を生成するしたがって 原因 方法などの関係を担う連用中止表現に対しても 適切な分割を実現することができる最後に評価実験によって 約86%の文に対して分割すべき述語を特定できること 約93%の連用中止表現に対して意味的な関係を2個以内に限定できること 約83%の文に対して有効な分割文を生成できることを示し 本方式の有効性を示す

文章作成において悪文を避け,わかりやすい文を作成するために多くのノウハウが提案されている.また,これらのノウハウを支援システムとして実現する試みも盛んに行われるようになってきた.しかし,文における表現の書換えまでを支援するシステムはまだ存在しない.われわれは,これらの推鼓上のノウハウを規則ベースとして持つことにより,表現の書換えまでを支援するシステムの研究開発を進めている本論文では,推敵を要しない文に対して高い計算コストをかけることが望ましくないという推蔽支援システムの特性を考慮し,構文多義をオンデマンドで解消しながら,規則として定義された部分的な書換えを段階的に行う過程により,書換え支援機能を実現する方式を提案する.このような書換え支援機能の実現形態においては,部分的な書換えを段階的に実行するための制御および,部分的な書換えに伴う内部データの整合性維持が必要となる.本論文で提案する方式においては,アジェンダを実行制御の基本的枠組みとし,その上に制約伝播に基づくデータ整合性維持機鶴を実現している.オンデマンドの構文多義解消と制約伝播によるデータ整合性維持を組み合わせることにより,必要なときに必要最小限の範囲で得た情報を最大限に利用することが可能となり,システム全体の処理効率の向上を図ることができる.

言語過程説の立場から 「対象」 「認識」 「表現」の関係に着目して (1)主体と客体に対する露者の認識の分析と (2)統語構造と意味の統一的扱い の2点が自然言語処理の重要な課題であることを示し これらの課題を実現する機械翻訳の方式として「多段翻訳方式」を提案する.本方式は上記の課題に対応する2つの部分的な方式 すなわち 主体的表現/客体的表現分離融合方式と多段変換方式から構成される.前者は原言語における主体的表現を解析し 話者の主観的感情や意志を抽出して目的言語に組み入れるものである.また後者は主体的表現情報抽出後の原文(客体的表現)を客体認識の繍造的抽象性のレベルに応じて特殊的認識構造 個別的認識構造 汎用的認識構造の3段階の構造的枠組みで捉え それぞれに対応する3つの変換方式 すなわち 慣用表現変換 意味的結合価パターン変換 汎用パターン変換によって目的言語に変換するものである.本方式は目的言語への変換の過程ではもちろん 原言語の解析の過程においても文構造の持つ意味をすくい取るものであるため 言語解研の基本的課題である多義解釈においても優れた効果が期待できる。また 変換規則は相互独立性が高く 相互矛盾の検証範囲が極小化されるため システムの成長が容易であると期待される.

日本語文の構造は 文の構文的・意味的な構造に大きく関わっている.よって 日本語文の処理を行う場合 その係り受け構造を明らかにすることが重要である.計算機により日本語文の係り受け解析を行う研究は広く行われているが そのために必要な知識の獲得・構造化を半自動的にかつ適応的に行うための研究は少ないように思われる.本論文では テキストデータに対する分析から直接的に以後の成長の核となる初期辞書データベースを構成する方法 および成長のための学習機能について検討した.この辞書データベース中では 単語はその係り受け特性に基づいてクラスタリングされ 分析対象としたテキスト中の単語間の係り受け関係は クラスタ間の係り受け可能関係として抽象化されて記述される.本辞書データベースは 実験文解析システムESSAYによる文解析に適用され その評価を受ける.さらに 解析が不成功である文において そのネガティブな状況より獲得される情報を用い 学習構造化の処理を受ける.このように 言語要素(単語等)の使われ方に基づいて知識の獲得を行うため 対象世界における拘束を緩やかに含んだ形の知識を得ることができる.また辞書データベースという記述的な形で構造化を行うため それ自身インクリメンタルに成長することが可能となった.

<p>We are developing a conversation system which efficiently transfers a massive amount of information like news articles by spoken dialogue. Here, "efficient" means that only the necessary information is transferred except unnecessary information for the user from target articles. In our system, feedbacks from the user are indispensable in order to realize high EoIT (Efficiency of Information Transfer). Therefore, we propose a utterance intention recognition method combining language information and prosodic information for the purpose of understanding diverse feedbacks from users. The feature of the proposed method is that it automatically extracts prosodic features with a high contribution ratio by using deep learning. We confirmed the effectiveness of the proposed method using a corpus with utterance intention tags designed based on dialogue data collected using our conversation system.</p>

<p>In a human-human conversation, a listener frequently conveys various intentions explicitly/implicitly to the speaker with reflexive short responses. The speaker recognizes the intentions of these feedbacks and changes the utterance plans to make the communication more smooth and efficient. These functions are expected to be useful for human-system conversation also, but when human face the system they do not always give the same feedback as they did with human. We investigated the feedback phenomena of human users against our system which is designed to transfer a massive amount of information like news articles by spoken dialogue. First, user's intentions of feedbacks that can affect the behavior of the system were classified from the viewpoint of roles in the progress of conversation: presence/absence of a user's interest, comprehension state, release/keep turn, and so on. Then, using our news delivery conversation system, we gathered user's short responses, and labeled the intention through listening test by the subjects. We also attempted automatic identification of those intentions. As a result, in the current system, it is found that there are many feedbacks to convey doubt and automatic identification of them is possible stably, but there are not so many other feedbacks.</p>

<p>想起は，ある概念が別の概念を思い浮かばせる程度と定義される．伝統的な語彙意味論に基づく意味関係を補完するものとして，多義解消や意味的類似度などの言語処理の応用への適用が想定されている．本研究では，想起を概念連鎖(連想)のショートカットであると仮定し，背景にある概念連鎖を意味ネットワーク上の経路として分析する．系列パターンマイニングの手法を適用することにより，想起の背景にある概念連鎖の類型を探る．</p>

<p>ニュース記事の伝達をタスクとして，聞き手が挟む相槌や聞き返しに応じて適宜情報を補完しながら会話を進める音声対話システムを提案する．要点を伝える主発話計画と，それに対し情報を補完する副発話計画を記事から自動生成する機能，これらの発話計画を聞き手の反応に暗に表れる理解状態に応じて切り替える機能を実装した．これらにより，リズムある対話による効率的な情報伝達が実現できた．</p>

This paper proposes a method to compute cross-lingual semantic textual similarities between sentences in different languages. The method is based on a machine-learning approach, which employs several monolingual similarity features chiefly proposed by projects participated in the *SEM STS (Semantic Textual Similarity) task. To utilize the monolingual features in a cross-lingual setting, we tested two methods to cross language borders: one utilizes bilingual dictionaries and the other employs off-the-shelf machine-translation engines. While our experimental results demonstrated such a translate then monolingual similarity method was reasonably effective in cross-lingual STS for English, Japanese, and Chinese, it has also been insisted that syntactic/structural differences have to be incorporated into the machine-learning based framework.

<p>文章中の語句に関連するWikipedia記事を対応付けるタスクはWikificationと呼ばれる．本発表は，日本語・英語の対訳文書に含まれる固有表現に対して並行的に対応付けを行う方式を提案する．提案方式は，他方の言語の情報を利用することにより，双方の言語における固有表現抽出処理の結果を補完し，言語横断的なWikificationを達成する．</p>

Wikipedia has attracted many researchers trying to extract structured information from its resources. In this research, we particularly draw attention on the multilingualism of Wikipedia and the semi-structured table-like information (Infobox) provided in each of the Wikipedia articles. More precisely, this research aims to establish a useful computational mechanism to extract multilingual attribute information from the Infobox templates and the actual Wikipedia articles in Japanese, Chinese, and English Wikipedias. The experimental results suggest that a clue, what we call attribute variables, plays a significant role in achieving high precision, and the overall performance can be further improved by incorporating attribute names and values.

This paper proposes a method to discover conceptual mates in the lexical-semantic resources of other languages. The proposed method integrates two types of cross-lingual semantic relatedness by balancing them using a weighted sum : One, definitional relatedness, estimates the probabilistic correspondence between sets of synonyms, while the other, explana-tory relatedness, measures the textual similarity between gloss texts. Our experimental results show that (1) the weighted-sum method was simple yet relatively robust; (2) the extended gloss texts gathered from the conceptual neighbors of objective concepts played a certain role in improving matching performances (mean average precision); and especially, (3) the nega-tive impact of machine translation, inevitably introduced to measure monolingual textual similarities, could be substantially reduced, and performances even slightly improved, thanks to the translation redundancy yielded by the use of multiple trans-lation engines.

This paper proposes a method for finding possibly corresponding synonym sets in a different lexical resource in a different language. By making use of Princeton Annotated Gloss Corpus as the source of target language statistics, the proposed method exhibited promising results in the experiments: More than 73% of the corresponding Princeton WordNet synsets were successfully recovered within the top-5 candidates, given a query set of Japanese WordNet synsets. As the proposed method minimally requires that the input to be seen as an apparently synonymous word set, the method could be extended and the performance would be further improved by incorporating richer information such as textual gloss overlaps and/or structural constraints posed by the lexical resources at hand.

In a Wab-based language service infrastructure, Web-servicized language resources can be combined and a new language resource is virtually realized as a composite Web service. Furthermore, if the infrastructure, as in the Language Grid, provides a collaborative environment, a new type of language resource that evolves through utilizations is feasible. This paper discusses possible architecture for a combined lexicon access service, which utilizes monolingual lexical semantic lexicons and bilingual dictionaries. This service could virtually realize a multilingual semantic lexicon by opportunistically associating semantically corresponding entries in the relevant lexical resources across languages. Given this scenario, this paper particularly proposes a framework for representing cross-lingual/interlingual lexical semantic correspondences that are expected to be incrementally recovered through a series of service invocations. This paper further discusses that recovered correspondences can be organized as a dynamic secondary language resource, while keeping the existing primary language resources intact.

We have been investigating characteristics of empathic states expressed in spontaneous utterances by analyzing speech data annotated with attitude and emotion categories. Basing on these investigations, this paper examines a method to estimate empathic states from speech utterances. The idea is to use the annotated data as training data for a machine learning algorithm (SVM) to construct relevant classifiers. These classifiers are arranged in such a way to reflect our definition of empathic states. The experimental results suggest: (1) there exist dependencies between attitude and emotion; (2) the estimation can be improved if augmented with the categories for the preceding utterance. This paper further discusses details of the initial experimental resutls and proposes a research agenda for the future study.

The purpose of this study was to examine how a transition of the conversation would occur among strangers, when they were to engage in multiple conversations and also to investigate how the transition would be associated with the atmosphere, and nonverbal behavior, such as the number of gesture, .head turning toward partner, the amount of utterance. In the experiment, 66 university students an unacquainted with one another participated as one of three-person group participated in dyadic and triadic conversations. Then, the influence of dyadic conversation on following triadic conversation was analyzed particularly in regard with the perceived activation level of atmosphere and nonverbal behavior.

The purpose of this study was to examine how a transition of the conversation would occur among strangers, when they were to engage in multiple conversations and also to investigate how the transition would be associated with the atmosphere, and nonverbal behavior, such as the number of gesture, .head turning toward partner, the amount of utterance. In the experiment, 66 university students an unacquainted with one another participated as one of three-person group participated in dyadic and triadic conversations. Then, the influence of dyadic conversation on following triadic conversation was analyzed particularly in regard with the perceived activation level of atmosphere and nonverbal behavior.

Aiming at computational estimation of participants' affective states in spontaneous dialogues, this paper analyzes the statistical relationships between affective states and speech features, such as F0 frequency, speech intensity, and speech speed. The results from statistical analyses applied to human-annotated speech data show that certain affective states, especially emotional states, are expressed and conveyed by particular speech features. The results further suggest that there exist possible gender differences in ways representing affective states.

The purpose of this study was to examine correlations among skill of building relationships, nonverbal behaviors (i.e. hand movements, interpersonal distance, head direction and the amount of utterance) and personality in a dyadic conversation. A total of 66 undergraduates (51 males and 15 females) participated in this experiment. The results were that skill of building relationships was positively correlated with ACT, self-monitoring and hand movements, and it was also negatively correlated with LSO-U. In addition, LSO-U was positively correlated with the amount of utterance.

From a Brunswikian lens model approach, this study examined relationships between nonverbal cues (i.e. hand movements, interpersonal distance, head direction and the amount of utterance) and skill of building relationships. A total of 66 undergraduates (51 males and 15 females) participated in this study. Participants rated their own skill of building relationships and were also rated by their conversational partners. Moreover, independent raters, who were not in this experiment, rated all participants on the skill. As a result, hand movements were perceived as a good indicator of the skill by those who actually exhibited as well as by their conversational partners and the independent raters.

This study experimentally investigated relationships between such nonverbal behaviors as degree of speech, hand gestures, head direction and body distance, and subjective evaluation of an activity of atmosphere created during a three-person conversation, in order to establish a technique for automatically estimating atmosphere in interpersonal communication. This study also investigated effect of information (topic) presentation by a display located in the conversation situation. Three participants who were the same generation and met for the first time talked for 8 minutes overall and then rated the activity of atmosphere of their own conversation presented by video every two minutes. Out results showed that the degree of speech and the hand gestures had a high correlation with subjective atmosphere while the head movement and the speech bias had a low correlation. Our results also revealed that the information presentation affected conversation and these correlations.

Language services provided by or registered to the Language Grid are classified into one of the predefined language service types, and can be employed through accordingly defined APIs. More than sixty actual Web services have been provided by the infrastructure, and the number is expected to further increase. In order to properly accommodate these services of various types, it is required to establish a principled way to organize them. Given this situation, this paper argues that a kind of ontological knowledge that backs up classification and definition of the services, including their APIs, has to be organized. In addition, this paper discusses role of the international standards associated with linguistic data and language resources. These standards should be highly appreciated not only for securing so-called interoperability, but also for characterizing some types of the language services. This paper further describes some experimental Web services whose results are represented in XML documents that are compliant with particular ISO standards : LAF/GrAF for syntactic dependency, and LMF for conceptual lexicon.

This paper discusses characteristics of empathic states expressed in spontaneous utterances. We basically define the empathic state of an utterance as a combination of attitude and emotional state that are expressed toward the preceding utterance to which the target utterance is directed. Among the possible combinations, we define &quot;empathy&quot; as a special empathic state, where the attitude and the emotional state are identical. Through a series of statistical analyses of speech data collected from three-party spontaneous dialogues, we show that some empathic states are identifiable by using a range of speech parameters. These results serve as solid foundation on which a computational mechanism for estimating a speaker&#039;s empathic states can be developed.

We propose an application independent Sinhala character input method called sri shell with a principled key assignment that is based on phonetic transcription of Sinhala characters. We argue that a good character input method should fulfill two criteria, which are efficiency and user-friendliness. We have introduced several quantification methods to quantify the efficiency and user-friendliness of Sinhala character input methods. Experimental results are shown to prove the efficiency and user-friendliness of the proposed method.

This paper proposes a method for generating Kanji reading alternatives for autonomous Kanji learning. The proposed method generates reading alternatives by applying rules based on pronunciation, shape, and semantic similarity of Kanjis. To reduce the number of alternatives properly, we applied an edit distance-based filter to the candidates generated through the pronunciation-based rules. This paper shows the preliminary evaluation results using the test questions formerly issued in the Japanese language proficiency tests, and discusses the effectiveness of the method.

To increase the accessibility and usability of online language services, this paper proposes the language grid to create composite language services for various communities. The language grid is called "horizontal," when the grid connects the standard languages of nations, or "vertical," when the grid combines the language services generated by communities. Semantic Web service technologies are applied in a human-centered fashion, to create composite language services through the collaboration of users and agents. Three example scenarios are given to illustrate how the language grid will organize standard and community language services for intercultural collaboration activities.

音声認識では認識辞書に含まれない単語は認識できないという、いわゆる未登録語の問題があるが、認識結果の内容に関連する語彙を獲得して認識辞書に追加することにより入力音声に適応した未登録語の少ない辞書を作成することができ、その辞書を用いて再度認識を行うことにより未登録語の影響を抑えて認識精度を改善できると考えられる。そこで本稿では、音声認識結果の内容に関連する語彙をコーパスから獲得し、獲得した語彙を辞書に追加することによって辞書を入力音声に適応させる手法を提案する。提案手法は、テキストコーパス中の語彙に対して語彙の分野を表す語彙分野ベクトルを算出しておき、入力の認識結果に対して発声内容の分野を推定し、その分野に近い語彙分野ベクトルを持つ語彙を入力に対する関連語彙として獲得し辞書に追加するという処理をオフラインで行う教師なし適応手法である。毎日新聞コーパスから各語彙の語彙分野ベクトルを求め、TVニュース音声を用いて提案手法の評価を行った。One of the most common problems in speech recognition system is the out-of-vocabulary (OOV) problem. Although we cannot avoid that input data includes OOV words for a vocabulary, we can reduce the number of them by adapting the vocabulary to inputs. Extracting relevant words to the content of an input speech based on a speech recognition result obtained using a reference vocabulary and adding them to the vocabulary enable to build an expanded vocabulary that includes less OOV words. The second recognition process using the new vocabulary is supposed to be performed better than the first process. In this paper, we propose vocabulary adaptation that acquires relevant words to an input from corpus and builds an expanded vocabulary by adding them to the reference vocabulary.

One of the most common problems in speech recognition system is the out-of-vocabulary (OOV) problem. Although we cannot avoid that input data includes OOV words for a vocabulary, we can reduce the number of them by adapting the vocabulary to inputs. Extracting relevant words to the content of an input speech based on a speech recognition result obtained using a reference vocabulary and adding them to the vocabulary enable to build an expanded vocabulary that includes less OOV words. The second recognition process using the new vocabulary is supposed to be performed better than the first process. In this paper, we propose vocabulary adaptation that acquires relevant words to an input from corpus and builds an expanded vocabulary by adding them to the reference vocabulary.

In this study, we analyzed user adaptation to machine translation(MT) systems through a controlled experiment. In the experiment, 67 subjects translated sentences using a Japanese-English-Japanese turnback translation, then we found out following results : 1) The more knowledge user has about source language, the better user can adapt to system, 2) The rule instruction beforehand affects for all user's adaptation, especially, it has a great effect for subjects who have medium level of source language, then following those who have low level, and those who have higher level.

In this study, we analyzed user adaptation to machine translation (MT) systems through a controlled experiment. In the experiment, 67 subjects translated sentences using a Japanese-English-Japanese turnback translation, then we found out following results: 1) The more knowledge user has about source language, the better user can adapt to system, 2) The rule instruction beforehand affects for all user's adaptation, especially, it has a great effect for subjects who have medium level of source language, then following those who have low level, and those who have higher level.

One of the most common problems in speech recognition system is the out-of-vocabulary (OOV) problem. Although we cannot avoid that input data includes OOV words for a vocabulary, we can reduce the number of them by adapting the vocabulary to inputs. Extracting relevant words to the content of an input speech based on a speech recognition result obtained using a reference vocabulary and adding them to the vocabulary enable to build an expanded vocabulary that includes less OOV words. The second recognition process using the new vocabulary is supposed to be performed better than the first process. In this paper, we propose vocabulary adaptation that acquires relevant words to an input from corpus and builds an expanded vocabulary by adding them to the reference vocabulary.

This paper describes an automatic multimedia content indexing system, which includes acoustic segmentation, automatic speech recognition, natural language processing, and video indexing features. The system mainly focuses on multimedia news programs. Speech segments, which are extracted from news content, are delivered to the speech recognition module. The speech recognition result sequence is segmented into topics, and those audio indexing results can be integrated with video indexing information to extract news story structure. Experiments show that the integrated results provide better news story structures than any single type of information

Recently XML has been considered as an emerging standard for describing resources on the Web. It is indispensable for next generation Web search systems to rank XML documents based on their relevance to the query and structural constraints.This paper describes requirements for earch systems, which handle XML documents describing resources on the Web. It then describes a new search system, which makes index files according to a format file specifying 1)the sub-structure of XML documents, 2)the type of index files and 3)the schema of extracting terms to satisfy the requirements.

Recently XML has been considered as an emerging standard for describing resources on the Web. It is indispensable for next generation Web search systems to rank XML documents based on their relevance to the query and stuructural constraints. This paper describes requirements for search systems, which handle XML documents describing resources on the Web. It then describes a new search system, which makes index files according to a format file specifying 1)the sub-structure of XML documents, 2)the type of index files and 3)the schema of extracting terms to satisfy the requirements.

The noun part of the WordNet 1.6 English lexical database was translated into Japanese by human translators. In the course of the translation work, elemental English words/collocations in a synset were translated into Japanese counterparts, as well as the "gloss", which defines the English lexical concept. This translation work turns out to give us an aligned and conceptually indexed pararell corpus. This paper shows the basic characteristics of the "Japanized" noun part of the WordNet. We argue that the resulted data might be a good resource for comparative linguistics. In addition to these, we examine how the derived bilingual lexical data can be applied to cross-linguistic natural language applications, such as cross-1anguage information retrieval.

This paper proposes an architecture for distrubuted cross-language informationo retrieval and a protocol used in it. The architecture includes metasearch engines and search engines, which communicate each other by using the protocol. The protocol is designed to communicate not only a search request and its result, but also inis also introduced, and it is concerned with formation of the functionalities and the contents provided by search engines The architecture allows a user to automatically choose and exploit various search engines probiding a large amount of documents in all sorts of languages in the Internet/WWW, according to his/her information need. The paper also introduced the joint project on a distributed cross-language information retrieval using the proposed architecture, involving KRDL in Singapore, KAIST in Korea and us NTT.

This paper proposes an architecture for distributed cross-language information retrieval and a protocol used in it. The architecture includes metasearch engines and search engines, which communicate each other by using the protocol. The protocol is designed to communicate not only a search request and its result, but also inis also introduced, and it is concerned with formation of the functionalities and the contents provided by search engines The architecture allows a user to automatically choose and exploit various search engines providing a large amount of documents in all sorts of languages in the Internet/WWW, according to his/her information need. The paper also introduces the joint project on a distributed cross-language information retrieval using the proposed architecture, involving KRDL in Singapore, KAIST in Korea and us NTT.

近年，電子化テキストの急激な増加，および，インターネットによる一般利用者の電子媒体への日常的なアクセスに伴って，言語処理研究と言語に関する実用技術の間のギャップが徐々に狭まってきており，実用的な自然言語処理研究という言葉が真に現実的な意味を持ち出してきた．本報告では，そのような実用的言語処理技術の事例のいくつかを「ここまでできるぞ言語処理技術」というタイトルで紹介する．In recent years, the amount of online text data rapidly increased, and ordinary users have come to access online texts through internet in their daily lives. This change around the situation of language data and language processing technologies gradually made the gap between research on natural language processing and practical technologies for processing natural language narrower, and has given the term &quot;reserach on practical natural language processing&quot; much more reality than before. This article reports case studies of those practical natural language processing technologies with the title &quot;Recent Advances in Language Processing Technologies&quot;.

In recent years, the amount of online text data rapidly increased, and ordinary users have come to access online texts through internet in their daily lives.This change around the situation of language data and language processing technologies gradually made the gap between research on natural language processing and practical technologies for processing natural language narrower, and has given the term "reserach on practical natural language processing" much more reality than before.This article reports case studies of those practical natural language processing technologies with the title "Recent Advances in Language Processing Technologies".

In recent years, the amount of online text data rapidly increased, and ordinary users have come to access online texts through internet in their daily lives. This change around the situation of language data and language processing technologies gradually made the gap between research on natural language processing and practical technologies for processing natural language narrower, and has given the term "reserach on practical natural language processing" much more reality than before. This article reports case studies of those practical natural language processing technologies with the title "Recent Advances in Language Processing Technologies".

近年, WWW上の検索エンジンが, 盛んに開発されている。このような検索エンジンは, 通常, 検索結果を検索要求と結果の類似度あるいは関連度に基づいて順位付けし出力する。本稿では, この検索結果を様々な観点 (ビュー)から再構成することにより, ナビゲーションを支援する機能について報告する。

World Wide Web(WWW)上の情報空間の拡大に伴い、いわゆる情報資源の探査の問題が顕著化し、WWW空間からの効率的な情報探索手法に関して議論が進められている。情報量が増大し、更に内容が多様化している情報空間の効率的なアクセスのためには、従来の情報探索手法はそのまま適用できない.本稿では、単なる情報空間の探索だけではなく情報空間の知的アクセス手法の実現を目指したインターネット情報探索システムTITAN (Total Information Traverse AgeNt)に関して述べる。

Recent progresses of the World-Wide Web (WWW) are remarkable. Already huge amount of information is available on the WWW, and it is expected to grow further. However, in general, it is not that easy for users to retrieve relevant information from the distributed information resources on the WWW. A user navigation processes on the WWW should consist of two sub processes; a process for locating relevant information resources based on their information needs, and that for browsing resources around the ones located initially. This paper first argues the resource discovery problem on the WWW, and proposes an incremental approach to it. A currently being developed prototype system, named TTTAN Ver.1, for the WWW navigation is then described. Finally, future research directions are discussed in terms of intelligent information processing including natural language processing.

Natural language processing is one of the most important technologies in the computerlized society. Special Interest Group of Natural Language Processing of IPSJ has been continuing its high activity since its establishment. In this paper we present the topics of the papers presented at its periodical conferences in this one decade, focusing on these three years. We also list a set of key words for the further elaboration in natural language research.

1992年7月23日から28日にかけて、第14回計算言語学国際会議 (e 14 th International Conferenceon Computational Linguistics，COLING?) がフランスのナント市会議場で開催された。COLINGは2年に1回開催されている計算言語学、自然言語処理関係の最大の国際会議である。今回は、500人以上の参加者を集め、約420件の投稿から採択された約200件の論文発表が行なわれた。本報告では、会議の概要および、いくつかの興味ある論文を紹介する。The 14th International Conference on Computational Linguistics (COLING-92) was held at City Congress Hall of Nantes, France from July 23rd to 28th. COLING is the largest international conference in the field of computational linguistics and natural language processing. Over 500 people from all over the world attended the conference and about 200 papers were presented. This report gives the outline of the conference and summarizes several papers, which the authors are interested in.

談話やテキストでは、2文間に何らかの関係を持ちながら展開していることが多く、その連鎖の仕方には対象世界の特徴が反映されていることが多い。これに基づき、日本語テキストの構造を、隣接する2文の並びのパターン（連接パターン）を基に解析する手法を提案する。ここでは、単付文のタイプとそれを接続する接続属性の組み合わせにより、文をタイプ分類し、そのタイプの並びのパターンから単位文をノードとするネットワーク構造を導出する手法を述べる。また、クローズドデータ、オープンデータによる評価実験の結果から、本手法の有効性、解析規則の収束性について考察する。This paper proposes a new method to analyze cohesive relations between adjacent Japanese sentences based on constraints naturally contained in objective texts. We especialy focused on explanative texts such as technical manuals. In these texts, the world to be explained constrains the sentence type and sentence adjacent patterns. For example, text for explaining a software command will describe some user operations and the associated system responses in some cohesive ways. We have developed a sentence type system suitable for software manuals and constructed a set of analysis rules. An experimental result toward the closed data shows the effectiveness of our method. We also discuss the coverage of the rule set through an experiment toward open data.

マニュアル文書を対象とした日本文推敲支援システムの研究開発を進めている。これは文章中の推敲すべき表現(不適性表現)を検出するとともに、それに対する書換え候補を提示するシステムである。マニュアルに頻出する不適性表現の1つに長文があり、上記システムにおいては、形態素解析情報に基づく文分割機能の検討を進めている。従来の文分割方式では、基本的に(1)分割する文節(以下、分割点と呼ぶ)の接続形式と分割時に挿入する接続詞を一対一に対応させる、(2)分割点を終止形に変換する、という方式が多い。このとき連用中止表現や接続助詞「て」による接続(以下、本稿では両者を併せて連用中止表現と呼ぶ)は、通常並列接続として扱われるため、上記のような方式では「並列の接続詞の挿入(または接続詞なし)、分割点文節の終止形への書換え」という分割方法を採ると考えられる。しかし、連用中止表現はその前後に特定の意味関係(以下、連用中止表現の意味用法という)を担うことがあることが指摘されている。また実際にマニュアル文中の連用中止表現を収集した結果でも、単なる並列接続とみなし得ないものが比較的多く存在する[figure]1(1)。さらに連用中止表現は、述語に付加される付属表現(時制表現や様相表現)が省略された表現形式であるため、分割点文節を終止形に変換するという方法では有効な分割文候補を生成することができない場合がある[figure]1(2)(3)。本稿では、文分割の観点から連用中止表現の意味用法を整理するとともに、表現の形態的特徴と意味用法、分割時の接続詞、連用中止文節の形態素操作との関係に着目した連用中止表現を分割点とする文の分割方式について述べる。

計算機による文書処理の研究開発が盛んに行われている。その中の一分野に、訂正支援あるいは推敲支援といわれる分野がある。成熟しつつある形態素解析技術を基盤として、単語レベルの訂正/推敲を支援するシステムがいくつか発表されている。しかし、文/文章レベルの推敲を支援するシステムは、あまり発表されていない。文/文章レベルの推敲支援を実現するためには、構文解析、意味解析などの言語解析の結果を用いてその中に含まれる推敲を心要とする表現(=不適性表現)を検出する技術をまず確立する必要がある。その際、不適性表現検出のための規則及びそれに付随する手続きをどのような枠組みで記述するか、また、その解釈をどのような機構によって行うかは、大きな問題である。 本稿では、特に文レベルの推敲支援を実現する際にその基本となる文レベル不適性表現の検出機構を提案する。

本稿では、日本語用言の持つ結合価構造を文解析の色々な局面に適用する手法について述べる。結合価構造は、結合価パターンと呼ぶ構造パターンとして辞書中に記述される。この構造パターンは、単位文の構造決定を行なうためのパターンマッチングに適用される他、埋め込み名詞句の構造決定、複合用言におけるゼロ代名詞化された格要素の補完、複合名詞・複合用言の解析などにも適用される。すなわち、結合価構造に基づくことにより、これらの言語現象に対しても統一的な解析を行なうことが可能となる。本稿では、さらに日英機械翻訳への適用における問題点についても論じる。In this paper, a method to apply the valency structure of Japanese predicates to Japanese sentences analysis is presented. In the dictionary, each of the valency structures is formulated as a sturucture pattern named &quot;Valency-pattern&quot;. This strutural pattern is applied to the analysis of unit-sentences, suppliment of zero-pronoun case-elements. analysis of embedded noun-phrases, analysis of complex words. These analysis processes are discussed. Taking this structure to be basis of the Japanese sentences analyzing process allows the uniform treatment of such language phenomena. The problems in applying this method to Japanese-to-English machine-translation are also discussed.

現在、開発中である日英翻訳システムALT-J/Eにおける日英変換技術について報告する。本システムの変換処理は、今回新たに提案する多段変換方式の考え方に基づいて設計されている。本稿では、多段変換方式の考え方、本システムにおける変換処理の位置付け・特徴、変換処理の概要及び、これに用いるパターン変換規則辞書の構成法について述べる。

現在、開発中の日英翻訳システムALT.-J/Eにおける日本語解析技術について報告する。本システムの日本語解析処理は、解析の各段階で生じる多義を効率的に絞り込むこと、多様な語いと表現をもつ文を取り扱えることに重点をおいて設計されている。以下、本稿では、上記のような点をいかにして達成するかということに的を絞り、本システムの日本語解析手法について述べる。

There are two methods for multilingualcommunication: using machine translation and using parallel texts. We studied the high-reliability of Back translation and parallel textsfor high-reliability multilingual communication. We also developed multilingual support software for medicalfield and education field.

Focusing on complex verbs, language resources, dictionaries, used in natural language processing were analyzed. Verbs in a dictionary used for morphological analysis were annotated by their morpho-semantic relations and structural relations, such as composition and deliberation. Meaning descriptions of complex verbs in a machine readable dictionary were analyzed from the perspective of their relations to the meanings of their components. These analyses provide some clues to organizing multiple language resources and providing those with a background of lexical conceptual structure theory.

It is crucially important to understand potential differences in concept lexicalization among various languages. To tackle this linguistic issue and to develop a computational mechanism for constructing a useful multilingual lexical ontology based on existing lexical resources, the reported research developed a method to associate possibly corresponding lexicalized concepts across languages by using a sense-tagged corpus, and proposed a framework for representing a dynamically acquired cross-lingual/ interlingual lexical-conceptual relations.