Search

SUMMARY To generate readable captions for Japanese spoken monologues such as lectures in real time, it is necessary to sequentially display captions that have proper linefeeds inserted. This paper proposes a technique for sequentially inserting proper linefeeds into a lecture transcript whenever a bunsetsu, which is a linguistic unit shorter than a sentence in Japanese and that roughly corresponds to a basic phrase in English, is identified. Under the assumption that linefeeds are inserted at bunsetsu boundaries, this technique can reduce the delay time of captioning to the utmost possible. This technique statistically judges whether or not a linefeed should be inserted into each bunsetsu boundary by using the information that is available at the time. We conducted experiments on linefeed insertion using a Japanese lecture corpus. The experimental results confirmed that our method, which is a bunsetsu-based linefeed insertion method, was almost as accurate as the sentence-based linefeed insertion method. In addition, we conducted comparative evaluations using four baseline methods. The results confirmed that our method could insert linefeeds more accurately than the simple methods that are thought to have the same delay time as our method.

Statutory sentences are generally difficult to read because of their complicated expressions and length. Such difficulty is one reason for the low quality of statistical machine translation (SMT). Multi-word expressions (MWEs) also complicate statutory sentences and extend their length. Therefore, we proposed a method that utilizes MWEs to improve the SMT system of statutory sentences. In our method, we extracted the monolingual MWEs from a parallel corpus, automatically acquired these translations based on the Dice coefficient, and integrated the extracted bilingual MWEs into an SMT system by the single-tokenization strategy. The experiment results with our SMT system using the proposed method significantly improved the translation quality. Although automatic translation equivalent acquisition using the Dice coefficient is not perfect, the best system’s score was close to a system that used bilingual MWEs whose equivalents are translated by hand.

The development of a captioning system that supports the real-time understanding of monologue speech such as lectures and commentaries is required. In monologues, since a sentence tends to be long, each sentence is often displayed in multi lines on the screen. In the case, it is necessary to insert linefeeds into a text so that the text becomes easy to read. This paper proposes a rule-based technique for inserting linefeeds into a Japanese spoken monologue sentence as an elemental technique to generate the readable captions. Our method inserts linefeeds into a sentence by applying the rules based on morphemes, dependencies and clause boundaries. We established the rules by circumstantially investigating the corpus annotated with linefeeds. An experiment using Japanese monologue corpus has shown the effectiveness of our rules.

It is very significant to the knowledge society to accumulate spoken documents on the web. However, because of the high redundancy of spontaneous speech, the faithfully transcribed text is not readable on an Internet browser, and therefore not suitable as a web document. This paper proposes a technique for converting spoken documents into web documents for the purpose of building a speech archiving system. The technique edits automatically transcribed texts and improves their readability on the browser. The readable text can be generated by applying technology such as paraphrasing, segmentation, and structuring transcribed texts. Editing experiments using lecture data demonstrated the feasibility of the technique. A prototype system of spoken document archiving was implemented to confirm its effectiveness.

Spoken monologues feature greater sentence length and structural complexity than spoken dialogues. To achieve high-parsing performance for spoken monologues, simplifying the structure by dividing a sentence into suitable language units could prove effective. This paper proposes a method for dependency parsing of Japanese spoken monologues based on sentence segmentation. In this method, dependency parsing is executed in two stages: at the clause level and the sentence level. First, dependencies within a clause are identified by dividing a sentence into clauses and executing stochastic dependency parsing for each clause. Next, dependencies across clause boundaries are identified stochastically, and the dependency structure of the entire sentence is thus completed. An experiment using a spoken monologue corpus shows the effectiveness of this method for efficient dependency parsing of Japanese monologue sentences.

Spontaneously spoken Japanese includes a lot of grammatically ill-formed linguistic phenomena such as fillers, hesitations, inversions, and so on, which do not appear in written language. This paper proposes a novel method of robust dependency parsing using a large-scale spoken language corpus, and evaluates the availability and robustness of the method using spontaneously spoken dialogue sentences. By utilizing stochastic information about the appearance of ill-formed phenomena, the method can robustly parse spoken Japanese including fillers, inversions, or dependencies over utterance units. Experimental results reveal that the parsing accuracy reached 87.0 ％, and we confirmed that it is effective to utilize the location information of a bunsetsu, and the distance information between bunsetsus as stochastic information.

Spontaneously spoken Japanese includes a lot of grammatically ill-formed linguistic phenomena such as fillers, hesitations, inversions, and so on, which do not appear in written language. This paper proposes a method of robust dependency parsing using a large-scale spoken language corpus, and evaluates the availability and robustness of the method using spontaneously spoken dialogue sentences. By utilizing stochastic information about the appearance of ill-formed phenomena, the method can robustly parse spoken Japanese including fillers, inversions, or dependencies over utterance units. As a result of an experiment, the parsing accuracy provided 87.0%, and we confirmed that it is effective to utilize the location information of a bunsetsu, and the distance information between bunsetsus as stochastic information., Grant-in-Aids for Young Scientists of the Ministry of Education, Science, Sports and Culture, Japan;The Tatematsu Foundation

Spontaneous speech includes a broad range of linguistic phenomena characteristic of spoken language, and therefore a statistical approach would be effective for robust parsing of spoken language. Though a largescale syntactically annotated corpus is required for the stochastic parsing, its construction requires a lot of human resources. This paper proposes a method of efficiently constructing a spoken language corpus for which the dependency analysis is provided. This method uses an existing spoken language corpus. A stochastic dependency parse is employed to tag spoken language sentences with the dependency structures, and the results are corrected manually. The tagged corpus is constructed in a spiral fashion where in the corrected data is utilized as the statistical information for automatic parsing of other data. Taking this spiral approach reduces the parsing errors, also allowing us to reduce the correction cost. An experiment using 10,995 Japanese utterances shows the spiral approach to be effective for efficient corpus construction.

E-mail systems are common communication tools, and it is desirable that e-mailmessages are written readably for recipients. One of techniques to write readable e-mail messages is to insert linefeeds and blank lines appropriately. However, linefeeds and blank lines in incoming e-mails are not always inserted at positions where recipients feel readable. If linefeeds and blank lines are inserted automatically at proper positions, readability of e-mail texts are improved and recipients can read e-mail texts efficiently. This paper proposes a method for text formatting of email texts by inserting linefeeds and blank lines into incoming e-mails at positions where recipients feel readable.

By appearance of social media, people are coming to be able to transmit information easily on a personal level. However, because users of social media generally spend little time on describing information, low-quality texts are transmitted and it blocks the spread of information. On transmitted texts in social media, commas and linefeeds are inserted incorrectly, and it becomes a factor of low-quality texts. This paper proposes a method for automatically formatting Japanese texts in social media. Our method formats texts by inserting commas and linefeeds appropriately. In our method, the positions where commas and linefeeds should be inserted are decided based on machine learning using morphological information, dependency relation and clause boundary information. An experiment using Japanese spoken language texts has shown the effectiveness of our method.

The development of a captioning system that supports the real-time understanding of monologue speech such as lectures and commentaries is required. In monologues, since a sentence tends to be long, each sentence is often displayed in multi lines on the screen and becomes unreadable. In the case, it is necessary to insert linefeeds into a text so that the text becomes easy to read. This paper proposes a technique for inserting linefeeds into a Japanese spoken monologue sentence as an elemental technique to generate the readable captions. Our method inserts linefeeds into a sentence by applying the rules based on morphemes, dependencies and clause boundaries. We established the rules by circumstantially investigating the corpus annotated with linefeeds. An experiment using Japanese monologue corpus has shown the effectiveness of our rules.

It is very significant in the knowledge society to accumulate spoken documents on the web. However, because of the high redundancy of spontaneous speech, the transcribed text in itself is not readable on an Internet browser, and therefore not suitable as a web document. This paper proposes a technique for converting spoken documents into web documents for the purpose of building a speech archiving system. The technique edits automatically transcribed texts and improves its readability on the browser. The readable text can be generated by applying technology such as paraphrasing, segmentation and structuring to the transcribed texts. An edit experiment using lecture data showed the feasibility of the technique. A prototype system of spoken document archiving was implemented to confirm its effectiveness.

To support the real-time understanding of spoken monologue such as lectures and commentaries, the development of a captioning system is required. In monologues, since a sentence tends to be long, each sentence is often displayed in multi lines on one screen, it is necessary to insert linefeeds into a text so that the text becomes easy to read. This paper proposes a technique for inserting linefeeds into a Japanese spoken monologue text as an elemental technique to generate the readable captions. Our method appropriately inserts linefeeds into a sentence by machine learning, based on the information such as dependencies, clause boundaries, pauses and line length. An experiment using Japanese speech data has shown the effectiveness of our technique., P09;1060

The development of a captioning system that supports the real-time understanding of monologue speech such as lectures and commentary is now in demand. In a realtime captioning system, it is necessary to summarize speech so that the audience can understand it within the display time and to output the caption simultaneously with the monologue speech input. This paper proposes a technique for simultaneous summarization of Japanese spoken monologue toward real-time captioning. Our technique identifies a unit for which the summarization is executed each time a clause boundary is detected. Then our technique summarizes it based on the dependency structure. An experiment using Japanese monologues has shown the feasibility of our technique.

Researchers of the CIAIR project at Nagoya University have constructed a data collection vehicle and have collected about 179 hours of multi-modal data. Speech data from about 800 subjects have been transcribed and speech intentions, dependency structures, and dialogue structures to the text data have been annotated. Various research activities within the project’s scope are continuing using the annotated data such as speech intention understanding and speaker’s knowledge acquisition. In this chapter, we introduce these research activities and present the several findings from the in-car speech corpus.

Spoken monologues feature greater sentence length and structural complexity than do spoken dialogues. To achieve high parsing performance for spoken monologues, it could prove effective to simplify the structure by dividing a sentence into suitable language units. This paper proposes a method for dependency parsing of Japanese monologues based on sentence segmentation. In this method, the dependency parsing is executed in two stages: at the clause level and the sentence level. First, the dependencies within a clause are identified by dividing a sentence into clauses and executing stochastic dependency parsing for each clause. Next, the dependencies over clause boundaries are identified stochastically, and the dependency structure of the entire sentence is thus completed. An experiment using a spoken monologue corpus shows this method to be effective for efficient dependency parsing of Japanese monologue sentences, P06;1022

同時通訳や字幕生成のように, 独話を同時的に処理する音声言語処理システムでは, 音声入力にしたがって順次, 解析を実行する漸進的解析技術が必要である.本論文では, 節を解析単位とする独話の漸進的係り受け解析手法を提案する.本手法では, 節境界解析に基づき, 話者による音声入力と同時的に節を同定する.節が入力されるたびにその節の内部の係り受け構造を作成し, すでに入力された節との係り受け関係を動的に決定する.独話文全体が入力される前の段階で係り受け関係を出力することが可能であり, 同時的な音声理解のための言語解析技術として利用できる.独話データを用いた解析実験により, 本手法が, 従来の独話文係り受け解析と同程度の解析性能を備えていることを確認した., In applications of spoken monologue processing such as simultaneous machine interpretation and automatic captions generation, incremental language parsing is strongly required. This paper proposes a technique for incremental dependency parsing of spoken Japanese monologue on a clause-by-clause basis. The technique identifies the clauses based on clause boundaries analysis, analyzes the dependency structures of them, and tries to decide the dependency relations with another clauses, simultaneously with the monologue speech input. The dependency relations are outputted at the stage before the input of the entire monologue sentence, and therefore, our technique can be used for language parsing in simultaneous Japanese speech understanding. An experiment using Japanese monologues has shown that our technique had the same degree of the performance as our past dependency parsing for monologue sentences.

This paper, which aims to increment the vocabulary of an existing thesaurus using hyponymy relations, focuses on an agricultural thesaurus called AGROVOC. Our main goal is to acquire AGROVOC-qualified candidates from the hyponymy relations of legal texts and tables. We propose a pattern-based approach to hyponymy relation acquisition. Our experimental result showed that 222 and 868 candidates are extracted from statutory sentences with 67.1% precision and tables with 37.0% precision, respectively.

This paper proposes a method for reordering words in a Japanese sentence based on concurrent execution with dependency parsing so that the sentence becomes more readable. Our contributions are summarized as follows: (1) we extend a probablistic model used in the previous work which concurrently performs word reordering and dependency parsing; (2) we conducted an evaluation experiment using our semi-automatically constructed evaluation data so that sentences in the data are more likely to be spontaneously written by natives than the automatically constructed evaluation data in the previous work.