scispace - formally typeset
Search or ask a question
Journal

Technical report of IEICE. Thought and language 

About: Technical report of IEICE. Thought and language is an academic journal. The journal publishes majorly in the area(s): Parsing & Sentence. Over the lifetime, 34 publications have been published receiving 82 citations.

Papers
More filters
Journal Article
TL;DR: Two event-related potentials to long locative adjuncts following different types of NPs are reported, suggesting that accusative objects impose a larger memory load than dative objects.
Abstract: We report event-related potentials to long locative adjuncts following different types of NPs. First, when the adjunct followed an accusative object NP, it elicited a sustained right anterior negativity compared to when it followed a nominative subject NP, suggesting that accusative objects impose a larger memory load. Second, after a dative object NP, the adjunct elicited a combination of N400 and sustained left anterior positivity compared to when it followed a nominative subject NP, suggesting that a larger mismatch cost is involved in the processing of dative objects. The results are compatible with a model that measures memory load based on the syntactic representations built incrementally by a subclass of left-corner parsers. In such a model, an accusative object NP leads to the prediction of a verb, therefore memory load increases when a VP-internal adjunct intervenes between the accusative object NP and the predicted verb. In contrast, a dative object NP leads to the prediction of an accusative NP; therefore, a mismatching cost is incurred when the prediction is falsified by a VP-internal adjunct. The nominative subject serves as a baseline because it predicts neither a verb nor a direct object. The results are not compatible with previous models such as the Dependency Locality Theory, which use dependency relations to predict memory load. Keyword: left-corner parsing, ERP, working memory, parsing algorithm, DLT, sentence processing, Japanese 日本語文処理における文処理方略とワーキングメモリー負荷の 事象関連電位(ERP)による研究 -left-corner parsing と Dependency Locality Theory の観点から内田 翔大 1,3 宮本 エジソン 正 2 広瀬 友紀 1 小林 由紀 伊藤 たかね 1 1 東京大学大学院 総合文化研究科 言語情報科学専攻 〒153-8902 東京都目黒区駒場 3-8-1 2 筑波大学人文社会科学系 〒305-8577 茨城県つくば市天王台 1-1-1 3 E-mail: shodai@phiz.c.u-tokyo.ac.jp 概要 この研究では、事象関連電位(ERP)実験を行い、文頭名詞句の格のみが異なる文の処理を比較した。まず、場所を表す 付加詞が文頭のヲ格目的語に続く場合、文頭のガ格主語に続く場合に対し、持続性右側前方陰性波が検出され、ヲ格目的語によ って大きなワーキングメモリー負荷があることを示す。次に、付加詞が文頭の二格目的語に続く場合、文頭のガ格主語に続く場 合に対し、N400 と持続性左側前方陽性波が検出され、二格目的語によって予測とは違う入力による処理負荷があることを示す。 この結果は left-corner parser のサブクラスによって累積的に構築される統語構造に基づいてワーキングメモリー負荷が計算され るモデルの予測に合致する。このようなモデルでは、ヲ格目的語の名詞句によって動詞が予測され、ヲ格目的語名詞句と予測さ れた動詞の間に VP 内付加詞が挿入されると、ワーキングメモリー負荷が増大する。また、二格目的語の名詞句によってヲ格目 的語が予測されるので、VP 内付加詞によって予測が裏切られると、予測に反した事による付加が増大する。一方、主格の名詞 句ではそのどちらの予測もなされないので基準条件の役割を果たす。この結果は Dependency Locality Theory のようなワーキン グメモリー負荷を統語的依存関係から説明しようとするような今までのモデルでは説明でないものであった。 キーワード Left-corner、事象関連電位、ワーキングメモリー、文処理方略、DLT、文処理、日本語

4 citations

Journal Article
TL;DR: A method is proposed by Center-Band obtained by using the result of the character recognition only statistically to improve the structure analysis of Eto and Suzuki's formula recognition.
Abstract: The mathematical formula recognition presented by “Eto and Suzuki” last year is robust against errors of the character recognition to some extent. However, there are some cases causing a fatal error of structure analysis. We propose a method by Center-Band obtained by using the result of the character recognition only statistically. It considerably improves the structure analysis. We report our new method and the result of the experiment.

4 citations

Network Information
Related Journals (5)
IEICE Transactions on Information and Systems
7.1K papers, 45.8K citations
75% related
Cognitive Science
7.2K papers, 231.3K citations
71% related
Journal of Psycholinguistic Research
1.9K papers, 52.2K citations
71% related
Speech Communication
2.6K papers, 119K citations
70% related
Language, cognition and neuroscience
908 papers, 13.8K citations
70% related
Performance
Metrics
No. of papers from the Journal in previous years
YearPapers
20153
20145
20131
20111
20081
20072